Determining The Chronology And Causality Of Events

Abstract

An example method includes calculating latency bounds for communications from two sensors to a collector (i.e., maximum and minimum latencies). After the collector receives an event report from the first sensor and an event report form the second sensor, the collector can determine, using the latency bounds, whether one event likely preceded the other.

Description

RELATED APPLICATIONS

[0001] This application claims priority to U.S. Provisional Application No. 62/171,899, entitled "SYSTEM FOR MONITORING AND MANAGING DATACENTERS", filed 5 Jun. 2015, which is incorporated herein by reference in its entirety.

TECHNICAL FIELD

[0002] The present technology pertains to network security and more specifically pertains to determining the chronology of events with the network.

BACKGROUND

[0003] Data centers typically include large numbers of entities (e.g., servers, switches, routers, etc.), each of which has its own internal clock used to annotate the time of events. These entities are rarely fully synchronized and, across the data center, their clocks often present substantial discrepancies. This is known as the clock skew problem.

[0004] Clock skew can make it difficult to determine event sequences (i.e., what is the proper chronological ordering of the events) and event causality (which events triggered which events). If all clocks in the data center were guaranteed to be in exact synchrony then ascertaining chronological ordering of events and event causality could be achieved by simply looking at event timestamps. However, since the various clocks in the network are not guaranteed to be in exact synchrony even when using clock management mechanisms such as NTP, determining event chronology and causality in the data center can prove difficult. This is particularly difficult when dealing with events that occurred very close in time.

BRIEF DESCRIPTION OF THE FIGURES

[0005] In order to describe the manner in which the above-recited and other advantages and features of the disclosure can be obtained, a more particular description of the principles briefly described above will be rendered by reference to specific embodiments that are illustrated in the appended drawings. Understanding that these drawings depict only example embodiments of the disclosure and are not therefore to be considered to be limiting of its scope, the principles herein are described and explained with additional specificity and detail through the use of the accompanying drawings in which:

[0006] FIG. 1 shows an example network traffic monitoring system according to some example embodiments;

[0007] FIG. 2 illustrates an example network environment 200 according to some example embodiments;

[0008] FIG. 3 shows an example method 300 for determining the chronological ordering of two events according to various embodiments;

[0009] FIG. 4A shows an example graphical representation of the timing of an event according to various embodiments;

[0010] FIG. 4B depicts an example timeline according to various embodiments;

[0011] FIGS. 5A, 5B, 5C, 5D, and 5E show example chronological graphs according to various embodiments; and

[0012] FIGS. 6A and 6B illustrate example system embodiments.

DESCRIPTION OF EXAMPLE EMBODIMENTS

Overview

[0013] The present technology includes determining precedence and causality of network events.

[0014] An example method includes calculating latency bounds for communications from two sensors to a collector (i.e., maximum and minimum latencies). After the collector receives an event report from the first sensor and an event report form the second sensor, the collector can determine, using the latency bounds, whether one event likely preceded the other.

Description

[0015] Various embodiments of the disclosure are discussed in detail below. While specific implementations are discussed, it should be understood that this is done for illustration purposes only. A person skilled in the relevant art will recognize that other components and configurations may be used without parting from the spirit and scope of the disclosure.

[0016] The disclosed technology addresses the need in the art for determining the chronology and causality of events in a data center.

[0017] Traditional data centers mainly contemplated attack vectors from outside the data center. Thus, firewalls and other security devices have traditionally been placed on the periphery of the data center. However, as modern data centers have begun to house diverse customers with a growing number of applications, internal security within a datacenter is becoming more important. Sophisticated attacks (or malfunctioning or misconfigured components) can be difficult to identify and trace. A crucial element of successfully identifying attacks can be to establish a trustworthy chronological history of events. The present disclosure enables a system to identify events with particularity across a datacenter and then piece together an accurate record. As discussed in greater detail below, this can be accomplished despite untrustworthy clocks on components or variable latency throughout the data center.

[0018] FIG. 1 shows an example network traffic monitoring system 100 according to some example embodiments. Network traffic monitoring system 100 can include configuration and image manager 102, sensors 104, external data sources 106, collectors 108, analytics module 110, policy engine 112, and presentation module 116. These modules may be implemented as hardware and/or software components. Although FIG. 1 illustrates an example configuration of the various components of network traffic monitoring system 100, those of skill in the art will understand that the components of network traffic monitoring system 100 or any system described herein can be configured in a number of different ways and can include any other type and number of components. For example, sensors 104 and collectors 108 can belong to one hardware and/or software module or multiple separate modules. Other modules can also be combined into fewer components and/or further divided into more components.

[0019] Configuration and image manager 102 can provision and maintain sensors 104. In some example embodiments, sensors 104 can reside within virtual machine images, and configuration and image manager 102 can be the component that also provisions virtual machine images.

[0020] Configuration and image manager 102 can configure and manage sensors 104. When a new virtual machine is instantiated or when an existing one is migrated, configuration and image manager 102 can provision and configure a new sensor on the machine. In some example embodiments configuration and image manager 102 can monitor the health of sensors 104. For instance, configuration and image manager 102 may request status updates or initiate tests. In some example embodiments, configuration and image manager 102 can also manage and provision virtual machines.

[0021] In some example embodiments, configuration and image manager 102 can verify and validate sensors 104. For example, sensors 104 can be provisioned a unique ID that is created using a one-way hash function of its basic input/output system (BIOS) universally unique identifier (UUID) and a secret key stored on configuration and image manager 102. This UUID can be a large number that is difficult for an imposter sensor to guess. In some example embodiments, configuration and image manager 102 can keep sensors 104 up to date by installing new versions of their software and applying patches. Configuration and image manager 102 can obtain these updates automatically from a local source or the Internet.

[0022] Sensors 104 can reside on nodes of a data center network (e.g., virtual partition, hypervisor, physical server, switch, router, gateway, other network device, other electronic device, etc.). In general, a virtual partition may be an instance of a virtual machine (VM) (e.g., VM 104a), sandbox, container (e.g., container 104c), or any other isolated environment that can have software operating within it. The software may include an operating system and application software. For software running within a virtual partition, the virtual partition may appear to be a distinct physical server. In some example embodiments, a hypervisor (e.g., hypervisor 104b) may be a native or "bare metal" hypervisor that runs directly on hardware, but that may alternatively run under host software executing on hardware. Sensors 104 can monitor communications to and from the nodes and report on environmental data related to the nodes (e.g., node IDs, statuses, etc.). Sensors 104 can send their records over a high-speed connection to collectors 108 for storage. Sensors 104 can comprise a piece of software (e.g., running on a VM, container, virtual switch, hypervisor, physical server, or other device), an application-specific integrated circuit (ASIC) (e.g., a component of a switch, gateway, router, standalone packet monitor, or other network device including a packet capture (PCAP) module or similar technology), or an independent unit (e.g., a device connected to a network device's monitoring port or a device connected in series along a main trunk of a datacenter). It should be understood that various software and hardware configurations can be used as sensors 104. Sensors 104 can be lightweight, thereby minimally impeding normal traffic and compute resources in a datacenter. Sensors 104 can "sniff" packets being sent over its host network interface card (NIC) or individual processes can be configured to report traffic to sensors 104. This sensor structure allows for robust capture of granular (i.e., specific) network traffic data from each hop of data transmission.

[0023] As sensors 104 capture communications, they can continuously send network traffic data to collectors 108. The network traffic data can relate to a packet, a collection of packets, a flow, a group of flows, etc. The network traffic data can also include other details such as the VM BIOS ID, sensor ID, associated process ID, associated process name, process user name, sensor private key, geo-location of a sensor, environmental details, etc. The network traffic data can include information describing the communication on all layers of the Open Systems Interconnection (OSI) model. For example, the network traffic data can include signal strength (if applicable), source/destination media access control (MAC) address, source/destination internet protocol (IP) address, protocol, port number, encryption data, requesting process, a sample packet, etc.

[0024] In some example embodiments, sensors 104 can preprocess network traffic data before sending to collectors 108. For example, sensors 104 can remove extraneous or duplicative data or they can create a summary of the data (e.g., latency, packets and bytes sent per flow, flagged abnormal activity, etc.). In some example embodiments, sensors 104 can be configured to only capture certain types of connection information and disregard the rest. Because it can be overwhelming for a system to capture every packet in a network, in some example embodiments, sensors 104 can be configured to capture only a representative sample of packets (e.g., every 1,000th packet or other suitable sample rate).

[0025] Sensors 104 can send network traffic data to one or multiple collectors 108. In some example embodiments, sensors 104 can be assigned to a primary collector and a secondary collector. In other example embodiments, sensors 104 are not assigned a collector, but can determine an optimal collector through a discovery process. Sensors 104 can change where they send their network traffic data if their environments change, such as if a certain collector experiences failure or if a sensor is migrated to a new location and becomes closer to a different collector. In some example embodiments, sensors 104 can send different types of network traffic data to different collectors. For example, sensors 104 can send network traffic data related to one type of process to one collector and network traffic data related to another type of process to another collector.

[0026] Collectors 108 can serve as a repository for the data recorded by sensors 104. In some example embodiments, collectors 108 can be directly connected to a top of rack switch. In other example embodiments, collectors 108 can be located near an end of row switch. Collectors 108 can be located on or off premises. It will be appreciated that the placement of collectors 108 can be optimized according to various priorities such as network capacity, cost, and system responsiveness. In some example embodiments, data storage of collectors 108 is located in an in-memory database, such as dashDB by International Business Machines. This approach benefits from rapid random access speeds that typically are required for analytics software. Alternatively, collectors 108 can utilize solid state drives, disk drives, magnetic tape drives, or a combination of the foregoing according to cost, responsiveness, and size requirements. Collectors 108 can utilize various database structures such as a normalized relational database or NoSQL database.

[0027] In some example embodiments, collectors 108 may only serve as network storage for network traffic monitoring system 100. In other example embodiments, collectors 108 can organize, summarize, and preprocess data. For example, collectors 108 can tabulate how often packets of certain sizes or types are transmitted from different nodes of a data center. Collectors 108 can also characterize the traffic flows going to and from various nodes. In some example embodiments, collectors 108 can match packets based on sequence numbers, thus identifying traffic flows and connection links. In some example embodiments, collectors 108 can flag anomalous data. Because it would be inefficient to retain all data indefinitely, in some example embodiments, collectors 108 can periodically replace detailed network traffic flow data with consolidated summaries. In this manner, collectors 108 can retain a complete dataset describing one period (e.g., the past minute or other suitable period of time), with a smaller dataset of another period (e.g., the previous 2-10 minutes or other suitable period of time), and progressively consolidate network traffic flow data of other periods of time (e.g., day, week, month, year, etc.). By organizing, summarizing, and preprocessing the network traffic flow data, collectors 108 can help network traffic monitoring system 100 scale efficiently. Although collectors 108 are generally referred to herein in the plurality, it will be appreciated that collectors 108 can be implemented using a single machine, especially for smaller datacenters.

[0028] In some example embodiments, collectors 108 can receive data from external data sources 106, such as security reports, white-lists (106a), IP watchlists (106b), whois data (106c), or out-of-band data, such as power status, temperature readings, etc.

[0029] In some example embodiments, network traffic monitoring system 100 can include a wide bandwidth connection between collectors 108 and analytics module 110. Analytics module 110 can include application dependency (ADM) module 160, reputation module 162, vulnerability module 164, malware detection module 166, etc., to accomplish various tasks with respect to the flow data collected by sensors 104 and stored in collectors 108. In some example embodiments, network traffic monitoring system 100 can automatically determine network topology. Using network traffic flow data captured by sensors 104, network traffic monitoring system 100 can determine the type of devices existing in the network (e.g., brand and model of switches, gateways, machines, etc.), physical locations (e.g., latitude and longitude, building, datacenter, room, row, rack, machine, etc.), interconnection type (e.g., 10 Gb Ethernet, fiber-optic, etc.), and network characteristics (e.g., bandwidth, latency, etc.). Automatically determining the network topology can assist with integration of network traffic monitoring system 100 within an already established datacenter. Furthermore, analytics module 110 can detect changes of network topology without the need of further configuration.

[0030] Analytics module 110 can determine dependencies of components within the network using ADM module 160. For example, if component A routinely sends data to component B but component B never sends data to component A, then analytics module 110 can determine that component B is dependent on component A, but A is likely not dependent on component B. If, however, component B also sends data to component A, then they are likely interdependent. These components can be processes, virtual machines, hypervisors, virtual local area networks (VLANs), etc. Once analytics module 110 has determined component dependencies, it can then form a component ("application") dependency map. This map can be instructive when analytics module 110 attempts to determine a root cause of a failure (because failure of one component can cascade and cause failure of its dependent components). This map can also assist analytics module 110 when attempting to predict what will happen if a component is taken offline. Additionally, analytics module 110 can associate edges of an application dependency map with expected latency, bandwidth, etc. for that individual edge.

[0031] Analytics module 110 can establish patterns and norms for component behavior. For example, it can determine that certain processes (when functioning normally) will only send a certain amount of traffic to a certain VM using a small set of ports. Analytics module can establish these norms by analyzing individual components or by analyzing data coming from similar components (e.g., VMs with similar configurations) Similarly, analytics module 110 can determine expectations for network operations. For example, it can determine the expected latency between two components, the expected throughput of a component, response times of a component, typical packet sizes, traffic flow signatures, etc. In some example embodiments, analytics module 110 can combine its dependency map with pattern analysis to create reaction expectations. For example, if traffic increases with one component, other components may predictably increase traffic in response (or latency, compute time, etc.).

[0032] In some example embodiments, analytics module 110 can use machine learning techniques to identify security threats to a network using malware detection module 166. For example, malware detection module 166 can be provided with examples of network states corresponding to an attack and network states corresponding to normal operation. Malware detection module 166 can then analyze network traffic flow data to recognize when the network is under attack. In some example embodiments, the network can operate within a trusted environment for a time so that analytics module 110 can establish baseline normalcy. In some example embodiments, analytics module 110 can contain a database of norms and expectations for various components. This database can incorporate data from sources external to the network (e.g., external sources 106). Analytics module 110 can then create access policies for how components can interact using policy engine 112. In some example embodiments, policies can be established external to network traffic monitoring system 100 and policy engine 112 can detect the policies and incorporate them into analytics module 110. A network administrator can manually tweak the policies. Policies can dynamically change and be conditional on events. These policies can be enforced by the components depending on a network control scheme implemented by a network. Policy engine 112 can maintain these policies and receive user input to change the policies.

[0033] Policy engine 112 can configure analytics module 110 to establish or maintain network policies. For example, policy engine 112 may specify that certain machines should not intercommunicate or that certain ports are restricted. A network and security policy controller (not shown) can set the parameters of policy engine 112. In some example embodiments, policy engine 112 can be accessible via presentation module 116. In some example embodiments, policy engine 112 can include policy data 112. In some example embodiments, policy data 112 can include endpoint group (EPG) data 114, which can include the mapping of EPGs to IP addresses and/or MAC addresses. In some example embodiments, policy data 112 can include policies for handling data packets.

[0034] In some example embodiments, analytics module 110 can simulate changes in the network. For example, analytics module 110 can simulate what may result if a machine is taken offline, if a connection is severed, or if a new policy is implemented. This type of simulation can provide a network administrator with greater information on what policies to implement. In some example embodiments, the simulation may serve as a feedback loop for policies. For example, there can be a policy that if certain policies would affect certain services (as predicted by the simulation) those policies should not be implemented. Analytics module 110 can use simulations to discover vulnerabilities in the datacenter. In some example embodiments, analytics module 110 can determine which services and components will be affected by a change in policy. Analytics module 110 can then take necessary actions to prepare those services and components for the change. For example, it can send a notification to administrators of those services and components, it can initiate a migration of the components, it can shut the components down, etc.

[0035] In some example embodiments, analytics module 110 can supplement its analysis by initiating synthetic traffic flows and synthetic attacks on the datacenter. These artificial actions can assist analytics module 110 in gathering data to enhance its model. In some example embodiments, these synthetic flows and synthetic attacks are used to verify the integrity of sensors 104, collectors 108, and analytics module 110. Over time, components may occasionally exhibit anomalous behavior. Analytics module 110 can analyze the frequency and severity of the anomalous behavior to determine a reputation score for the component using reputation module 162. Analytics module 110 can use the reputation score of a component to selectively enforce policies. For example, if a component has a high reputation score, the component may be assigned a more permissive policy or more permissive policies; while if the component frequently violates (or attempts to violate) its relevant policy or policies, its reputation score may be lowered and the component may be subject to a stricter policy or stricter policies. Reputation module 162 can correlate observed reputation score with characteristics of a component. For example, a particular virtual machine with a particular configuration may be more prone to misconfiguration and receive a lower reputation score. When a new component is placed in the network, analytics module 110 can assign a starting reputation score similar to the scores of similarly configured components. The expected reputation score for a given component configuration can be sourced outside of the datacenter. A network administrator can be presented with expected reputation scores for various components before installation, thus assisting the network administrator in choosing components and configurations that will result in high reputation scores.

[0036] Some anomalous behavior can be indicative of a misconfigured component or a malicious attack. Certain attacks may be easy to detect if they originate outside of the datacenter, but can prove difficult to detect and isolate if they originate from within the datacenter. One such attack could be a distributed denial of service (DDOS) where a component or group of components attempt to overwhelm another component with spurious transmissions and requests. Detecting an attack or other anomalous network traffic can be accomplished by comparing the expected network conditions with actual network conditions. For example, if a traffic flow varies from its historical signature (packet size, transport control protocol header options, etc.) it may be an attack.

[0037] In some cases, a traffic flow may be expected to be reported by a sensor, but the sensor may fail to report it. This situation could be an indication that the sensor has failed or become compromised. By comparing the network traffic flow data from multiple sensors 104 spread throughout the datacenter, analytics module 110 can determine if a certain sensor is failing to report a particular traffic flow.

[0038] Presentation module 116 can include serving layer 118, authentication module 120, web front end 122, public alert module 124, and third party tools 126. In some example embodiments, presentation module 116 can provide an external interface for network monitoring system 100. Using presentation module 116, a network administrator, external software, etc. can receive data pertaining to network monitoring system 100 via a webpage, application programming interface (API), audiovisual queues, etc. In some example embodiments, presentation module 116 can preprocess and/or summarize data for external presentation. In some example embodiments, presentation module 116 can generate a webpage. As analytics module 110 processes network traffic flow data and generates analytic data, the analytic data may not be in a human-readable form or it may be too large for an administrator to navigate. Presentation module 116 can take the analytic data generated by analytics module 110 and further summarize, filter, and organize the analytic data as well as create intuitive presentations of the analytic data.

[0039] Serving layer 118 can be the interface between presentation module 116 and analytics module 110. As analytics module 110 generates reports, predictions, and conclusions, serving layer 118 can summarize, filter, and organize the information that comes from analytics module 110. In some example embodiments, serving layer 118 can also request raw data from a sensor or collector.

[0040] Web frontend 122 can connect with serving layer 118 to present the data from serving layer 118 in a webpage. For example, web frontend 122 can present the data in bar charts, core charts, tree maps, acyclic dependency maps, line graphs, tables, etc. Web frontend 122 can be configured to allow a user to "drill down" on information sets to get a filtered data representation specific to the item the user wishes to drill down to. For example, individual traffic flows, components, etc. Web frontend 122 can also be configured to allow a user to filter by search. This search filter can use natural language processing to analyze the user's input. There can be options to view data relative to the current second, minute, hour, day, etc. Web frontend 122 can allow a network administrator to view traffic flows, application dependency maps, network topology, etc.

[0041] In some example embodiments, web frontend 122 may be solely configured to present information. In other example embodiments, web frontend 122 can receive inputs from a network administrator to configure network traffic monitoring system 100 or components of the datacenter. These instructions can be passed through serving layer 118 to be sent to configuration and image manager 102 or policy engine 112. Authentication module 120 can verify the identity and privileges of users. In some example embodiments, authentication module 120 can grant network administrators different rights from other users according to established policies.

[0042] Public alert module 124 can identify network conditions that satisfy specified criteria and push alerts to third party tools 126. Public alert module 124 can use analytic data generated or accessible through analytics module 110. One example of third party tools 126 is a security information and event management system (SIEM). Third party tools 126 may retrieve information from serving layer 118 through an API and present the information according to the SIEM's user interfaces.

[0043] FIG. 2 illustrates an example network environment 200 according to some example embodiments. It should be understood that, for the network environment 100 and any environment discussed herein, there can be additional or fewer nodes, devices, links, networks, or components in similar or alternative configurations. Example embodiments with different numbers and/or types of clients, networks, nodes, cloud components, servers, software components, devices, virtual or physical resources, configurations, topologies, services, appliances, deployments, or network devices are also contemplated herein. Further, network environment 200 can include any number or type of resources, which can be accessed and utilized by clients or tenants. The illustrations and examples provided herein are for clarity and simplicity.

[0044] Network environment 200 can include network fabric 212, layer 2 (L2) network 206, layer 3 (L3) network 208, endpoints 210a, 210b, . . . , and 210d (collectively, "204"). Network fabric 212 can include spine switches 202a, 202b, . . . , 202n (collectively, "202") connected to leaf switches 204a, 204b, 204c, . . . , 204n (collectively, "204"). Spine switches 202 can connect to leaf switches 204 in network fabric 212. Leaf switches 204 can include access ports (or non-fabric ports) and fabric ports. Fabric ports can provide uplinks to spine switches 202, while access ports can provide connectivity for devices, hosts, endpoints, VMs, or other electronic devices (e.g., endpoints 204), internal networks (e.g., L2 network 206), or external networks (e.g., L3 network 208).

[0045] Leaf switches 204 can reside at the edge of network fabric 212, and can thus represent the physical network edge. In some cases, leaf switches 204 can be top-of-rack switches configured according to a top-of-rack architecture. In other cases, leaf switches 204 can be aggregation switches in any particular topology, such as end-of-row or middle-of-row topologies. Leaf switches 204 can also represent aggregation switches, for example.

[0046] Network connectivity in network fabric 212 can flow through leaf switches 204. Here, leaf switches 204 can provide servers, resources, VMs, or other electronic devices (e.g., endpoints 210), internal networks (e.g., L2 network 206), or external networks (e.g., L3 network 208), access to network fabric 212, and can connect leaf switches 204 to each other. In some example embodiments, leaf switches 204 can connect endpoint groups (EPGs) to network fabric 212, internal networks (e.g., L2 network 206), and/or any external networks (e.g., L3 network 208). EPGs can be used in network environment 200 for mapping applications to the network. In particular, EPGs can use a grouping of application endpoints in the network to apply connectivity and policy to the group of applications. EPGs can act as a container for buckets or collections of applications, or application components, and tiers for implementing forwarding and policy logic. EPGs also allow separation of network policy, security, and forwarding from addressing by instead using logical application boundaries. For example, each EPG can connect to network fabric 212 via leaf switches 204.

[0047] Endpoints 210 can connect to network fabric 212 via leaf switches 204. For example, endpoints 210a and 210b can connect directly to leaf switch 204a, which can connect endpoints 210a and 210b to network fabric 212 and/or any other one of leaf switches 204. Endpoints 210c and 210d can connect to leaf switch 204b via L2 network 206. Endpoints 210c and 210d and L2 network 206 are examples of LANs. LANs can connect nodes over dedicated private communications links located in the same general physical location, such as a building or campus.

[0048] Wide area network (WAN) 212 can connect to leaf switches 204c or 204d via L3 network 208. WANs can connect geographically dispersed nodes over long-distance communications links, such as common carrier telephone lines, optical lightpaths, synchronous optical networks (SONET), or synchronous digital hierarchy (SDH) links. LANs and WANs can include layer 2 (L2) and/or layer 3 (L3) networks and endpoints.

[0049] The Internet is an example of a WAN that connects disparate networks throughout the world, providing global communication between nodes on various networks. The nodes typically communicate over the network by exchanging discrete frames or packets of data according to predefined protocols, such as the Transmission Control Protocol/Internet Protocol (TCP/IP). In this context, a protocol can refer to a set of rules defining how the nodes interact with each other. Computer networks may be further interconnected by an intermediate network node, such as a router, to extend the effective size of each network. Endpoints 210 can include any communication device or component, such as a computer, server, hypervisor, virtual machine, container, process (e.g., running on a virtual machine), switch, router, gateway, host, device, external network, etc. In some example embodiments, endpoints 210 can include a server, hypervisor, process, or switch configured with virtual tunnel endpoint (VTEP) functionality which connects an overlay network with network fabric 212. The overlay network may allow virtual networks to be created and layered over a physical network infrastructure. Overlay network protocols, such as Virtual Extensible LAN (VXLAN), Network Virtualization using Generic Routing Encapsulation (NVGRE), Network Virtualization Overlays (NVO3), and Stateless Transport Tunneling (STT), can provide a traffic encapsulation scheme which allows network traffic to be carried across L2 and L3 networks over a logical tunnel. Such logical tunnels can be originated and terminated through VTEPs. The overlay network can host physical devices, such as servers, applications, endpoint groups, virtual segments, virtual workloads, etc. In addition, endpoints 210 can host virtual workload(s), clusters, and applications or services, which can connect with network fabric 212 or any other device or network, including an internal or external network. For example, endpoints 210 can host, or connect to, a cluster of load balancers or an EPG of various applications.

[0050] Network environment 200 can also integrate a network traffic monitoring system, such as the one shown in FIG. 1. For example, as shown in FIG. 2, the network traffic monitoring system can include sensors 104a, 104b, . . . , 104n (collectively, "104"), collectors 108a, 108b, . . . 108n (collectively, "108"), and analytics module 110. In some example embodiments, spine switches 202 do not have sensors 104. Analytics module 110 can receive and process network traffic data collected by collectors 108 and detected by sensors 104 placed on nodes located throughout network environment 200. In some example embodiments, analytics module 110 can be implemented in an active-standby model to ensure high availability, with a first analytics module functioning in a primary role and a second analytics module functioning in a secondary role. If the first analytics module fails, the second analytics module can take over control. Although analytics module 110 is shown to be a standalone network appliance in FIG. 2, it will be appreciated that analytics module 110 can also be implemented as a VM image that can be distributed onto a VM, a cluster of VMs, a software as a service (SaaS), or other suitable distribution model in various other example embodiments. In some example embodiments, sensors 104 can run on endpoints 210, leaf switches 204, spine switches 202, in-between network elements (e.g., sensor 104h), etc. In some example embodiments, leaf switches 204 can each have an associated collector 108. For example, if leaf switch 204 is a top of rack switch then each rack can contain an assigned collector 108.

[0051] Although network fabric 212 is illustrated and described herein as an example leaf-spine architecture, one of ordinary skill in the art will readily recognize that the subject technology can be implemented based on any network topology, including any data center or cloud network fabric. Indeed, other architectures, designs, infrastructures, and variations are contemplated herein. For example, the principles disclosed herein are applicable to topologies including three-tier (including core, aggregation, and access levels), fat tree, mesh, bus, hub and spoke, etc. It should be understood that sensors and collectors can be placed throughout the network as appropriate according to various architectures.

[0052] FIG. 3 shows an example method 300 for determining the chronological ordering of two events according to various embodiments. Example method 300 begins with determining a minimum and maximum latency associated with a first sensor (step 302). The minimum latency and the maximum latency between collector 108 and sensor 104 can be calculated by one of sensor 104 or collector 108 sending a message to the other device (collector 108 or sensor 104) and calculating the time it takes for a response. The latency of the connection can be half the difference between the time the message was sent and the time the message was received. In some embodiments, over the course of normal operation collector 108 sends messages to sensor 104 and can calculate the minimum latency and maximum latency based on the latency of those messages and responses. Alternatively or additionally, collector 108 can send special messages such as a "ping" (i.e., an Internet Control Message Protocol echo request) specifically to calculate a minimum latency and maximum latency. In some embodiments, the minimum latency is assumed to be 0. In some embodiments, network traffic monitoring system 100 receives an input specifying a minimum latency for a connection to a sensor 104 or a default minimum latency for all connections. In some embodiments, the maximum and minimum latency are provided by sensor 104. In some embodiments, the latencies incorporate allowances for sensor 104 processing (i.e., if sensor 104 must process packets before sending a report to collector 108).

[0053] In some embodiments, the minimum latency and/or the maximum latency are estimated using or informed by a network graph which contains known or estimated latencies between network components. For example, without experimentally determining the minimum latency or the maximum latency, network traffic monitoring system 100 can estimate those latencies by estimating a path that a flow would take between collector 108 and sensor 104 and can, using previously determined the latencies of individual links within the flow, determine estimated minimum latency and maximum latency of the flow.

[0054] In some embodiments, the maximum latency is informed by varying timestamp to collector 108 received time differences. For example, if an event report with a timestamp of 30 (arbitrary time units) is first received at collector 108 time of 31 and then a later event report with a timestamp of 50 is received at collector 108 time of 61, then the maximum latency can be at least 10 ([61-50]-[31-30]). This maximum latency value can be useful even if the associated timestamp values are not trusted for global accuracy.

[0055] It should be understood that "latency" as used herein includes network latency as well as processing delays and other delays caused by physical or arbitrary limitations.

[0056] Using similar techniques as explained above in step 302, example method 300 can continue by determining a minimum and maximum latency associated with a second sensor (step 304).

[0057] Example method 300 can continue by receiving, at a first time and from a first sensor, a first event report describing a first event (step 306). An event can be a flow event such as a flow being initiated, received, rejected, ended, etc. An event can also include other data center, endpoint, or application, events (such as an process starting or freezing, an endpoint opening a port, a power outage, a bandwidth spike, an attack, etc.). An event can be a summary of other events. For example, in a DDoS situation, numerous attacking endpoints might attempt to communicate with a target endpoint; a sensor on each of the attacking endpoints can send an event report describing the various DDoS communications; a summary event can then be created describing that a DDoS attack occurred in the network.

[0058] An event report can include a summary of the event or a detailed report of an event. A summary of an event can be preprocessed by a sensor 104 to remove extraneous information or include extra (e.g., summary) information. In some embodiments, the event report includes a timestamp of the event, according to associated sensor's 104 clock. In some embodiments, sensor 104 sends reports to collector 108 immediately after the event occurs; alternatively, sensor 104 sends reports to collector 108 periodically and the reports can describe multiple events that occurred over the period.

[0059] Similar to step 306, Example method 300 can continue by receiving, at a second time and from a second sensor, a second event report describing a second event (step 308).

[0060] Example method 300 can continue by determining whether the first event caused the second event (step 310). For example, if the first event is a packet being sent from an endpoint associated with the first sensor and the second event is that same packet being received from an endpoint associated with the second sensor, then the first event caused the second event. The causal relation between the two events can be determined using other means as well; for example, analytics module 110 can determine that the first event was a trigger for the second event as might occur when the second event is a DDoS attack and the first event is the command and control message initiating the attack. If step 310 is "no", example method 300 can continue at step 312 and, similar to step 310, determine whether the second event caused the first event.

[0061] Step 314 follows a negative result at step 312 by determining if the first time minus the minimum latency for the first sensor is earlier than the second time minus the maximum latency for the second sensor. In some embodiments, this includes converting the times and latencies to integer values and determining which value is greater (i.e., earlier). Step 320 can follow a "no" result at step 314 and can include, similar to step 314, determining whether the second time minus the minimum latency for the second sensor is earlier than the first time minus the maximum latency for the first sensor.

[0062] In some embodiments, the system can attempt to find that one event (e.g., the first event) likely caused the another event (e.g., the second event) by determining that the times of the events are sufficiently close (e.g., within calculated or experimentally determined sensor-to-sensor latency bounds). The system can further determine that one event likely caused the other if the two events are complimentary in type; for example, if the first event is a flow being initialized (or a packet, frame, stream, etc. being sent) and the second event is a flow being received (or a packet, frame, stream, etc. being received). This can be useful if a packet or flow is translated (or otherwise obfuscated) as it traverses the network.

[0063] If a first event precedes a second event, in some embodiments, this means that the second event did not cause the first event. For example, if the first event is an attack, and the system is attempting to find the initial cause of the attack (e.g., a command and control signal), then if the second event is subsequent to the attack, then it likely did not cause the attack.

[0064] Following a negative result at step 320, network traffic monitoring system 100 can determine that the chronology between the first event and the second event is ambiguous (step 322). In other words, it cannot be determined based on the data available that either event preceded the other or that the events were simultaneous. In some embodiments, instead of having maximum and minimum latencies, the system can determine probability density functions for the latencies associated with the first and second sensors. Network traffic monitoring system 100 can then determine, using the two probability density functions, the likelihood that one event preceded the other.

[0065] If either step 310 or step 314 result in "yes," then example method 300 can continue by determining that the first event preceded the second event (step 316); similarly, if either step 312 or step 320 result in "yes," then example method 300 can continue by determining that the second event preceded the first event (step 318). After either step 316 or step 318, example method 300 can then end.

[0066] In some embodiments, after determining that the first event preceded the second event; the system can then determine whether the first event likely caused the second event. For example, the system can determine if the first time minus the maximum latency of the first sensor is close to the second time minus the minimum latency of the second sensor. In other words, the system can determine if the maximum possible difference between the actual times that the two events occurred is within a predetermined threshold. Other statistical models can be utilized to make a determination that the two events likely occurred within a predefined time window. In some embodiments, "causation" between events requires a different window based on the type of events. For example, in some attacks, the causal event should be within a narrow window (e.g., a few milliseconds) whereas in other events, the causal event might be within a broader window (e.g., a few hours).

[0067] In some embodiments, the system can also determine whether the one event caused another event by comparing the types of events of the two events. For example, if one event is sending a message and the other event is receiving the message (or a message that fits a description of the first message, although a direct match might be uncertain). Other examples of corresponding types of events can include a command and control signal being sent and an attack commencing, a flow being started and the flow being rejected, etc.

[0068] In some embodiments, the first event and the second event can be simultaneous. For example, the events can be receiving a timing signal from a universal clock (e.g., over a dedicated timing interface) received by the first and second sensors.

[0069] Although example method 300 concerns events from two different sensors 104; various techniques can be used to determine the chronology of two events from one sensor. For example, event reports can include a sequence number such that an event report with a later sequence number describes a later event in comparison to an event described in an event report with a lower sequence number. Another example includes where event reports include a timestamp of the events. Although the event timestamps might not necessarily be trusted, they can be used to determine the relative chronology for two events detected by the same sensor 104. Similarly, the time of receipt of the report by the collector 108 can inform the chronological ordering of two events from the same sensor, even if such ordering can be unreliable as one report may be delayed in transit. In some embodiments, a combination of techniques can be combined to determine a likely chronological ordering of events from a single senor 104 or a combination of sensors 108.

[0070] In some embodiments, a collector 108 can determine a sensor's 104 estimated clock skew based on a timestamp on a received communication, the collector's 108 clock when the communication is received, and the maximum and minimum latencies between the sensor 104 and collector 108. Example equation for this can be: Skew.sub.upperBound=-Time.sub.collector+Time.sub.sensor+Latency.sub.max and Skew.sub.lowerBound=-Time.sub.collector+Time.sub.sensor+Latency.sub.m- in. Applying these equations to an example, if the collector 108 receives a communication at time 76 (using arbitrary units of time for simplicity of explanation) bearing a timestamp of 51 and the latency between the collector 108 and sensor 104 varies from a maximum of 15 to a minimum of 3, then the clock skew is likely from -10 to -22 relative to the collector's 108 clock. Thus, a later communication from the same sensor 104 bearing a timestamp of 48 (sensor 104 time) can be translated to 58-70 (collector 108 time). This approach can be useful when receiving delayed event reports from sensors 104; for example, if a sensor 104 gathers event data over a period and then sends a report describing the events that happened during that period.

[0071] FIG. 4A shows an example graphical representation of the timing of an event 401 according to various embodiments and used in later figures. In some embodiments, network traffic monitoring system 100 can determine event window 402 based on the report received time 408 (i.e., at collector 108), minimum latency 406, and maximum latency 407. Event window 302 can be the estimated time (e.g., according to collector's 108 clock) calculated based on a minimum latency 306 and a maximum latency 407. For example, the earliest event time 403 can be calculated by subtracting the maximum latency 407 from report received time 408 while the latest event time 404 can be calculated by subtracting the minimum latency 406 from the report received time 408. Line 410 indicates the relative chronology of occurrences, occurrences to the right are more recent than occurrences to the left.

[0072] FIG. 4B depicts an example timeline 420 according to various embodiments. Various events 401 can correspond to sensors reporting from various endpoints. For example, events 401.sub.a1, 401.sub.a2, and 401.sub.a3 can correspond to Endpoint A; events 401.sub.b1, 401.sub.b2, and 401.sub.b3 can correspond to Endpoint B; and events 401.sub.c1, 401.sub.c2, and 401.sub.c3 can correspond to Endpoint C.

[0073] Referring to the event windows of event 401.sub.a1 and event 401.sub.a2, the latest event time of event 401.sub.a1 precedes the earliest event time of event 401.sub.a2. Thus, network monitoring system 100 can determine that event 401.sub.a1 preceded event 401.sub.a2 (see steps 314 and 316). In contrast, the latest time of event 401.sub.b1 is after the earliest time of event 401.sub.b2 while the latest time of event 401.sub.b2 is after the earliest time of event 401.sub.b1 (in other words, their event windows 402 overlap); thus network monitoring system 100 can determine that the chronology of the two events is ambiguous (see steps 314, 320, and 322). Applying the principles in example method 300, network monitoring system 100 can determine that event 401.sub.a1 precedes event 401.sub.a2 which both precede event 401.sub.a3.

[0074] FIGS. 5A-5E show example chronological graphs 501-505 according to various embodiments. Nodes A1-A3, B1-B3, and C1-C3 can correspond to respective events 401.sub.a1-401.sub.a3, 401.sub.b1-401.sub.b3, and 401.sub.c1-401.sub.c3 of FIG. 4B while nodes D1-D3 and E1-E3 correspond to events 401 not otherwise depicted. In FIGS. 5A-5E, an arrow can denote chorological precedence; e.g., that node A1 precedes node A2 in graph 501. Chorological precedence can be determined by applying example method 300 on various events 401 or through other means. A double arrow such as exists between nodes A2 and B2 in graph 502 can represent simultaneity (denoted herein with "="). Simultaneity can occur, for example, if two events 401 are in fact duplicate entries for one event (e.g., if received on two different collectors 108 or described in different contexts) or by using a dedicated timing interface.

[0075] Graphs 501-504 can represent various graph segments (e.g., from various collectors 108, methods, time periods, etc.). Graphs 501-505 can be directed acyclic graphs, meaning they represent strict precedence. For example in graph 501, because A1 precedes A2 (denoted using ">" herein) and A2>B1>B2>B3, then B3 cannot precede A1. Alternatively, the chronological precedence of events 408 can be probabilistic, which can result in a probability that one event 408 might precede another 408. For example, B3 might precede A1 in graph 501, according to the combined probabilities that each association is incorrect in the chain of A1>A2>B1>B2>B3, or some other statistical calculation.

[0076] Graphs can be combined; for example, graphs 501-504 can be combined to form graph 505. In some embodiments, the chronological precedence of two graphs may disagree. For example, graph 501 represents that A2>B1>B2 while graph 502 represents that A2=B2; this situation creates a "cycle" (A2>B1>B2=A2>B1 . . . ). Various conflict-resolution techniques can be employed to overcome such inconsistencies. For example, network traffic monitoring system 100 can determine probabilities that relationships are accurate and compare the probabilities; for example, A2>B1 in graph 501 might have a 99% confidence level while A2=B2 in graph 502 might have a 20% confidence level, the system can then disregard the A2=B2 association. In some embodiments, every relationship of the cycle can be disregarded (e.g., each of A2>B1, B1>B2, and A2=B2).

[0077] FIG. 6A and FIG. 6B illustrate example system embodiments. The more appropriate embodiment will be apparent to those of ordinary skill in the art when practicing the present technology. Persons of ordinary skill in the art will also readily appreciate that other system embodiments are possible.

[0078] FIG. 6A illustrates a conventional system bus computing system architecture 600 wherein the components of the system are in electrical communication with each other using a bus 605. Example system 600 includes a processing unit (CPU or processor) 610 and a system bus 605 that couples various system components including the system memory 615, such as read only memory (ROM) 670 and random access memory (RAM) 675, to the processor 610. The system 600 can include a cache of high-speed memory connected directly with, in close proximity to, or integrated as part of the processor 610. The system 600 can copy data from the memory 615 and/or the storage device 630 to the cache 612 for quick access by the processor 610. In this way, the cache can provide a performance boost that avoids processor 610 delays while waiting for data. These and other modules can control or be configured to control the processor 610 to perform various actions. Other system memory 615 may be available for use as well. The memory 615 can include multiple different types of memory with different performance characteristics. The processor 610 can include any general purpose processor and a hardware module or software module, such as module 1 637, module 7 634, and module 3 636 stored in storage device 630, configured to control the processor 910 as well as a special-purpose processor where software instructions are incorporated into the actual processor design. The processor 610 may essentially be a completely self-contained computing system, containing multiple cores or processors, a bus, memory controller, cache, etc. A multi-core processor may be symmetric or asymmetric.

[0079] To enable user interaction with the computing device 600, an input device 645 can represent any number of input mechanisms, such as a microphone for speech, a touch-sensitive screen for gesture or graphical input, keyboard, mouse, motion input, speech and so forth. An output device 635 can also be one or more of a number of output mechanisms known to those of skill in the art. In some instances, multimodal systems can enable a user to provide multiple types of input to communicate with the computing device 600. The communications interface 640 can generally govern and manage the user input and system output. There is no restriction on operating on any particular hardware arrangement and therefore the basic features here may easily be substituted for improved hardware or firmware arrangements as they are developed.

[0080] Storage device 630 is a non-volatile memory and can be a hard disk or other types of computer readable media which can store data that are accessible by a computer, such as magnetic cassettes, flash memory cards, solid state memory devices, digital versatile disks, cartridges, random access memories (RAMs) 675, read only memory (ROM) 670, and hybrids thereof.

[0081] The storage device 630 can include software modules 637, 634, 636 for controlling the processor 610. Other hardware or software modules are contemplated. The storage device 630 can be connected to the system bus 605. In one aspect, a hardware module that performs a particular function can include the software component stored in a computer-readable medium in connection with the necessary hardware components, such as the processor 610, bus 605, display 635, and so forth, to carry out the function.

[0082] FIG. 6B illustrates an example computer system 650 having a chipset architecture that can be used in executing the described method and generating and displaying a graphical user interface (GUI). Computer system 650 is an example of computer hardware, software, and firmware that can be used to implement the disclosed technology. System 650 can include a processor 655, representative of any number of physically and/or logically distinct resources capable of executing software, firmware, and hardware configured to perform identified computations. Processor 655 can communicate with a chipset 660 that can control input to and output from processor 655. In this example, chipset 660 outputs information to output 665, such as a display, and can read and write information to storage device 670, which can include magnetic media, and solid state media, for example. Chipset 660 can also read data from and write data to RAM 675. A bridge 680 for interfacing with a variety of user interface components 685 can be provided for interfacing with chipset 660. Such user interface components 685 can include a keyboard, a microphone, touch detection and processing circuitry, a pointing device, such as a mouse, and so on. In general, inputs to system 650 can come from any of a variety of sources, machine generated and/or human generated.

[0083] Chipset 660 can also interface with one or more communication interfaces 690 that can have different physical interfaces. Such communication interfaces can include interfaces for wired and wireless local area networks, for broadband wireless networks, as well as personal area networks. Some applications of the methods for generating, displaying, and using the GUI disclosed herein can include receiving ordered datasets over the physical interface or be generated by the machine itself by processor 655 analyzing data stored in storage 670 or 675. Further, the machine can receive inputs from a user via user interface components 685 and execute appropriate functions, such as browsing functions by interpreting these inputs using processor 655.

[0084] It can be appreciated that example systems 600 and 650 can have more than one processor 610 or be part of a group or cluster of computing devices networked together to provide greater processing capability.

[0085] For clarity of explanation, in some instances the present technology may be presented as including individual functional blocks including functional blocks comprising devices, device components, steps or routines in a method embodied in software, or combinations of hardware and software.

[0086] In some embodiments the computer-readable storage devices, mediums, and memories can include a cable or wireless signal containing a bit stream and the like. However, when mentioned, non-transitory computer-readable storage media expressly exclude media such as energy, carrier signals, electromagnetic waves, and signals per se.

[0087] Methods according to the above-described examples can be implemented using computer-executable instructions that are stored or otherwise available from computer readable media. Such instructions can comprise, for example, instructions and data which cause or otherwise configure a general purpose computer, special purpose computer, or special purpose processing device to perform a certain function or group of functions. Portions of computer resources used can be accessible over a network. The computer executable instructions may be, for example, binaries, intermediate format instructions such as assembly language, firmware, or source code. Examples of computer-readable media that may be used to store instructions, information used, and/or information created during methods according to described examples include magnetic or optical disks, flash memory, USB devices provided with non-volatile memory, networked storage devices, and so on.

[0088] Devices implementing methods according to these disclosures can comprise hardware, firmware and/or software, and can take any of a variety of form factors. Typical examples of such form factors include laptops, smart phones, small form factor personal computers, personal digital assistants, rackmount devices, standalone devices, and so on. Functionality described herein also can be embodied in peripherals or add-in cards. Such functionality can also be implemented on a circuit board among different chips or different processes executing in a single device, by way of further example.

[0089] The instructions, media for conveying such instructions, computing resources for executing them, and other structures for supporting such computing resources are means for providing the functions described in these disclosures.

[0090] Although a variety of examples and other information was used to explain aspects within the scope of the appended claims, no limitation of the claims should be implied based on particular features or arrangements in such examples, as one of ordinary skill would be able to use these examples to derive a wide variety of implementations. Further and although some subject matter may have been described in language specific to examples of structural features and/or method steps, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to these described features or acts. For example, such functionality can be distributed differently or performed in components other than those identified herein. Rather, the described features and steps are disclosed as examples of components of systems and methods within the scope of the appended claims. Moreover, claim language reciting "at least one of" a set indicates that one member of the set or multiple members of the set satisfy the claim.

Claims

1. A computer-implemented method comprising: receiving, from a first sensor at a first time, a description of a first event; receiving, from a second sensor at a second time, a description of a second event; and determining that the first event preceded the second event.

2. The computer-implemented method of claim 1, further comprising: determining a minimum latency associated with the first sensor; determining a maximum latency associated with the second sensor; and wherein the determining that the first event preceded the second event comprises: comparing the first time minus the minimum latency for the first sensor with the second time minus the maximum latency for the second sensor.

3. The computer-implemented method of claim 2, further comprising: determining a maximum latency associated with the first sensor; determining a minimum latency associated with the second sensor; and determining that the first event caused the second event by determining that the first time minus the maximum latency associated with the first sensor is within a predetermined range with the second time minus the minimum latency associated with the second sensor.

4. The computer-implemented method of claim 3, wherein determining that the first event caused the second event is based on the type of event of the first event and the type of event of the second event.

5. The computer-implemented method of claim 4, wherein the second event is a data center attack.

6. The computer-implemented method of claim 1, further comprising: receiving, from the first sensor, a description of a third event; receiving, from the second sensor, a description of a fourth event; wherein the determining that the first event preceded the second event comprises: determining that the first event preceded the third event; determining that the third event preceded the fourth event; and determining that the fourth event preceded the second event.

7. The computer-implemented method of claim 6, wherein: determining that the first event preceded the third event comprises comparing: a timestamp associated with the first event with a timestamp associated with the third event; or a sequence number associated with the first event with a sequence number associated with the third event; and determining that the third event preceded the fourth event comprises comparing: an event type associated with the third event with an event type associated with the fourth event.

8. The computer-implemented method of claim 7, wherein the first event is a data center attack, the method further comprising: determining that the second event did not cause the first event.

9. The computer-implemented method of claim 1, wherein the second event is a process starting.

10. The computer-implemented method of claim 1, wherein the second time is before the first time.

11. A non-transitory computer-readable medium having computer readable instructions stored thereon that, when executed by a processor of a computer, cause the computer to: receive, from a first sensor at a first time, a description of a first event; receive, from the first sensor, a description of a third event; receive, from a second sensor at a second time, a description of a second event; receive, from the second sensor, a description of a fourth event; and determine that the first event preceded the second event by: determining that the first event preceded the third event; determining that the third event preceded the fourth event; and determining that the fourth event preceded the second event.

12. The non-transitory computer-readable medium of claim 11, wherein: the instructions that cause the computer to determine that the first event preceded the third event further cause the computer to compare: a timestamp associated with the first event with a timestamp associated with the third event; or a sequence number associated with the first event with a sequence number associated with the third event; and the instructions that cause the computer to determine that the third event preceded the fourth event further cause the computer to compare an event type associated with the third event with an event type associated with the fourth event.

13. The non-transitory computer-readable medium of claim 12, wherein the first event is a data center attack and the instructions further cause the computer to: determine that the second event did not cause the first event.

14. The computer-implemented method of claim 1, wherein the second sensor is installed on a virtual machine and the second event is a process starting.

15. The computer-implemented method of claim 1, wherein the second time is before the first time.

16. A system comprising: a processor; a computer-readable medium; and non-transitory computer-readable instructions stored thereon that, when executed by the processor, cause the system to: determine a minimum latency associated with a first sensor; receive, from the first sensor at a first time, a description of a first event; determine a maximum latency associated with a second sensor; receive, from the second sensor at a second time, a description of a second event; and determine that the first event preceded the second event by: comparing the first time minus the minimum latency for the first sensor with the second time minus the maximum latency for the second sensor.

17. The system of claim 16, wherein the instructions further cause the system to: determine a maximum latency associated with the first sensor; determine a minimum latency associated with the second sensor; and determine that the first event caused the second event by determining that the first time minus the maximum latency associated with the first sensor is within a predetermined range with the second time minus the minimum latency associated with the second sensor.

18. The system of claim 17, wherein the instructions that cause the system to determine that the first event caused the second event do so based on the type of event of the first event and the type of event of the second event.

19. The system of claim 18, wherein the second event is a data center attack.

20. The system of claim 16, wherein the sensor is installed on a virtual machine and the second event is a process starting up.