Lime Brokerage has completed a series of initiatives and technology upgrades to improve and monitor latency of its trading and data systems, officials say.
In the fourth quarter of last year, Lime rolled out latency management technology from Dublin, Ireland-based Corvil, which is now in full production for monitoring latency across its trading servers, says Lime chief technology officer Suresh Thesayi.
The rollout was driven by the need to understand latencies across Lime’s network and servers. “We have always had a need for this, but a suitable solution was not available in a cost-effective manner,” Thesayi says. “The data collected can pinpoint bottlenecks in our Mesh network and pressure points during peak utilization. It also provides a nice correlation of latency against throughput.”
Initially, Lime will use the latency data internally to optimize the performance of its servers and network, and has plans to use it to offer performance dashboards to clients in the future, though he says a date for this is still to be decided.
However, the investment is already yielding results. Though the broker is still collecting data across various applications, the initial results showed “significant performance degredation at higher percentiles on older operating systems,” Thesayi says. As a result, Lime has since upgraded its trading servers to a new operating system, and has already seen “much better latency distribution,” he adds.
Currently, Lime monitors trading system latency and data from Nasdaq’s Inet platform in real time, and monitors the latency of other datafeeds on an ad-hoc basis due to the “bandwidth limitations” of a single 40 Gigabit-per-second monitoring device. However, Thesayi says this is temporary and that Lime will deploy another Corvil appliance to monitor latency of all datafeeds—though he adds that the broker hasn’t yet set a date for deploying an additional Corvil appliance.
In addition, Thesayi is currently awaiting internal approval for a hardware upgrade to Lime’s trading and data servers that will reduce one-way latency from around 12 microseconds to between five and six microseconds. “There is always a need for speed in the space Lime operated in. Cutting down the time it takes for a client to react to a change in the market is critical for the success of some strategies. Also, some clients and strategies are more sensitive to jitter than raw latency sometimes. These needs lead us to have a system that not only performs fast, but also uniformly,” he says.
The latest initiatives follow a “complete redesign” of Lime’s C++-based server-side market data application that provides processing of raw quotes and book-building to increase throughput and reduce latency, which began in 2012 and was completed in 2014 for equities feeds and mid-2015 for options feeds.
“The previous design had some performance bottlenecks with book processing, and had a much higher latency profile. So the primary force for the redesign is improved performance,” Thesayi says. Though unable to give specific figures for improvements achieved as a result because the Corvil platform was not yet in place, he says minimum latencies saw “little to no change,” but that “average and median latencies improved significantly since we were able to use more processing cores in the servers after the redesign.”
Lime is also working on several more improvements designed to “significantly” increase performance in 2016, he adds, though most of these are “too proprietary to share.”
For original version, please click here.