Corvil » Why CorvilNet?

Why CorvilNet?

The Corvil product has successfully addressed a number of the difficult challenges encountered by organizations attempting to do microsecond precision latency measurement at scale. These include:

Microsecond Precision Time Synchronization Over The Wide Area

Before any latency measurements can be made, the problem of microsecond precision time synchronization over the wide area must be addressed. Traditionally, the answer to this problem requires an external time synchronization signal to be provided to every latency measurement appliance. GPS is the most frequently used external time sync signal. The problem with this approach is that is expensive and very difficult to administer over large-scale deployments. In particular, when a trading infrastructure involves technology and services from multiple parties, it is often impossible to get a homogeneous time synchronization solution deployed.

While CorvilNet supports external clock synchronization from sources such as GPS, it also offers a unique automated time synchronization technology that delivers microsecond accuracy. The Corvil time synchronization technology requires zero external resources and is set up automatically between CorvilNet appliances once latency measurement channels are configured. This technology has been operating successfully in production environments since 2007 and is the time synchronization method of choice with Corvil customers. No other solution on the market offers this capability.

Assured Integrity of Underlying Latency Measurements

Unfortunately many people are becoming familiar with this problem. A typical deployment model for latency monitoring involves the use of SPAN ports on routers and switches. Of late, the use of active aggregation TAPs has also become popular. The problem with SPAN ports and aggregation TAPS is that they can become temporarily overloaded due to spikes in the traffic known as microbursts. These microbursts cause congestion and distort the latency profile of the underlying traffic before it reaches the latency measurement appliance. As a result the latency management appliance will give a false and inaccurate reading. This is often referred to as the GARBAGE IN = GARBAGE OUT problem.

Within the CorvilNet solution there is a SPAN/TAP Port Congestion Detection feature that automatically detects and alarms on any congestion occurring on the measurement port prior to the data being received. CorvilNet will flag any impacted measurements whose integrity cannot be guaranteed. Assuring the integrity and accuracy of any underlying latency management systems is of paramount importance and something, which is unique to the CorvilNet solution at the present time.

Distributed Latency Management Architecture Enabling Enterprise Scale

In building a Latency Management System one has to fundamentally decide from the beginning the scope and scale of the target solution. Is it primarily aimed at point deployments (e.g. within a co-lo center) or is it expected to cover enterprise scale deployments (e.g. within and between multiple data centers, co-lo centers and liquidity venues). Many of the solutions on the market use a centralized latency measurement architecture. Therefore, it is necessary to backhaul 100% of all captured packets and associated timestamps over the network to a central data store before time correlation and latency measurements can be made. This is extremely bandwidth and storage intensive and means that these solutions cannot be used over the WAN and often encounter significant scaling issues in enterprise deployments.

CorvilNet uses a distributed latency management architecture. There is no backhaul of packet captures from the measurement appliances to a central data store. Time correlation for latency measurement is achieved using a lightweight distributed protocol, which adds minimum overhead to the system. All measurements and analysis are done in real-time at the point of measurement. CorvilNet does however support centralized management via its Latency Management Center, which provides a single point of monitoring and configuration for all deployed CorvilNet appliances while preserving the scalable distributed nature of the underlying architecture. The distributed architecture approach is the main reason why the Corvil solution has been successful in Enterprise scale deployments.

Unified Application and Network Latency Management

The CorvilNet solution is the only offering on the market that provides integrated packet capture, network analysis and application analysis (i.e. transaction, message and middleware) in one single platform. Some solutions require a separate vendor (i.e. three vendors) for each of packet capture, network analysis and application analysis functions with no integration between them. For example, with Corvil it is possible to track latency of any FIX order per customer and compare the contributions to overall latency of the FIX gateway versus the network and then export all associated packet captures to an external platform for archiving or further analysis. All of this can be done in a single view with full time synchronization maintained between analytics at each layer. Likewise a user can look at the latency impact of a microburst at the network layer and determine what application transactions, messages, and clients were impacted.

Corvil offers decoder libraries for protocol analysis of the leading exchange and liquidity venue formats. In addition, a Decoder Library SDK is being introduced in June 2010 allowing customers to develop custom and/or proprietary decoders for protocols not currently supported in the Corvil native library.

Leader in Inter Party Latency Management

Corvil introduced the first Inter Party Latency Management Solution for Financial Markets in April 2009. This is called CorvilClear. Since then much discussion, debate and industry activity has happened surrounding Inter Party Latency (IPL) and the associated topic of latency transparency in the trading loop. Following the success of the Corvil initiative, it was proposed that a standards process would be undertaken by the FIX Protocol group. This is known as FIPL (FIX Inter Party Latency). Two models for Inter Party Latency measurement are emerging:

• MASTER-SLAVE IPL – this is where a latency provider party (i.e. the master) controls all aspects of the latency measurement and then elects to issue reports periodically to the latency consuming party (i.e. the slave).
• PEER-TO-PEER (P2P) IPL – this where any two parties agree to mutually exchange their latency information electronically in real-time with full transparency. In this case the parties are both providers and consumers of latency information.

The Corvil solution is unique in its ability to support both models and continues to be the only company that has successfully deployed both solutions in the market place:

• Deutsche Börse – is an example of the Master-Slave IPL Model where it issues detailed end of day reports on latency for order execution and market data via its client portal
• London Stock Exchange (Turquoise) – is an example of the P2P IPL Model where market participants can peer with the CorvilClear publishers hosted by the market center to receive continuous real-time exchange of latency information

In addition Corvil has established CorvilClear partnerships with leading service providers to the Financial Markets. An example of this is Savvis who is offering both a managed and unmanaged CorvilClear service using the P2P IPL Model for its clients. Savvis is hosting CorvilClear publishers at the leading equities, futures and options and FX market centers for US and EU markets. Therefore clients connected to the Savvis network can receive in real-time, microsecond precision latency monitoring, troubleshooting and reporting to and from all of these market centers.

Corvil continues to pioneer the development of IPL solutions and is active in the FIPL working group. Corvil intends to be fully compliant with and interoperable with any industry-recognized standard in this area.

Broad Range of Innovative Latency Analytics

Corvil has a proven track record for continuously adding new analytics to CorvilNet that leverages the company’s strong heritage in mathematics and scientific innovation. Such analytics include:

• PREDICTIVE LATENCY MEASUREMENT – algorithm that measures and computes the expected latency induced by microbursts destined to access links of a specified bandwidth. This is computed in real-time, Example: Predictive latency measurement can be used to alarm and trap on all microbursts detected that are expected to induce at least 2ms of latency on a 100Mb/s access link from the exchange to the market participant.
• RELATIVE LATENCY MEASUREMENT – ability to measure the relative latency between any two events happening in the trading network to microsecond precision without the need for external time synchronization. Events can be specified at the transaction, message and packet level.  Example: Relative latency measurement can be used to determine how much faster market price is being updated on a direct feed to the co-lo site compared to the consolidated feed at the central data center?
• MARKET DATA GAP DETECTION – ability to detect and report on gaps in the sequence numbers of received market data feeds. These gaps can cause significant latency problems in downstream processing of the market data stream and it is critical to determine where, and why they are occurring. Corvil provides a decoder library for all popular feed formats in US and EU markets.
• STATISTICAL SLA FRAMEWORK – ability to specify and report latency service levels compliance in a fashion appropriate for trading applications. Example: The framework can be used to specify and track latency SLAs (Service Level Agreements) of the form; for this market data feed, latency during the busy 1 second of the trading day should be no worse than 3.2ms for 99.9% of the traffic.
• BANDWIDTH REQUIREMENT FOR LATENCY AND LOSS – algorithm that computes in real-time the specific bandwidth needed to meet a statistical defined latency and loss SLAs. Example. This feature can be used to determine the bandwidth needed such that no more than 250 microseconds of latency is added to 99.99% of order entry traffic sent to BATS.
• SHAPED PEAK RATE – most implementations of peak rate measurement for traffic are inaccurate when the traffic has been previously shaped by an upstream connection. For example if traffic has come from a 100Mb/s circuit put passed through a lower bandwidth choke point along the path, then most implementations report the peak rate of the traffic as 100Mb/s. The answer should of course be the bandwidth of the choke point. Corvil provides a special feature for measuring the true shaped peak rate of traffic. This has often proved invaluable in detecting bandwidth choke points upstream from the point of observation.

Industry Leading Performance

Corvil has the highest performing latency management appliances on the market. The CNE 5400 (4 x 10Gb/s) is capable of processing over 5 million pps with 1μs granular real-time analysis. The CorvilNet Engines use specialized hardware time-stamping with nanosecond precision and performs all analytics in real-time before moving the packets to disk.

The performance of the CorvilNet engine is assisted through the use of the latest and fastest multi-core Intel processors combined with a multi-threaded 64-bit software architecture used for all CorvilNet applications.

Most other solutions move the packets to disk immediately before analysis is possible. This means that 100% of packets must always be captured and stored before analysis is possible. This often introduces a lag in reported information and makes live reporting very difficult. Any packets lost in the process can invalidate the analysis results.