Pico Leverages Latest 3rd Gen Intel Xeon Scalable Processors to reach 100G Real-time Analytics For Financial Markets

The Corvil Analytics Product from Pico has been the market leader in Trading and Market Data Analytics for many years. It is used by most of the large banks, stock exchanges, and brokers to manage and analyze their trading and market data plant. It’s a high-speed, high-scale, ultra-demanding environment, where every message and every microsecond absolutely counts. And each year, the market data rates and performance demands from our clients grow, and our product offerings need to keep ahead of this.

At the outset, we decided to base our appliances on the Intel platform, a decision we have doubled down on with each new Intel architecture release – we’re on the 6th or 7th generation of appliances now! Every couple of years Intel releases a new architecture, which we adopt, and this allows us to scale up our offering to meet the latest market demands. You may not be aware of this, but with Covid and lockdown, market data rates surged way beyond their expected growth curve, putting extra pressure on the financial markets infrastructure to perform.

Which made the arrival of the new latest Ice Lake 3rd Gen Intel Xeon Scalable processors architecture very timely indeed.

We got our first 3rd Gen Intel Xeon Scalable processor Ice Lake Silver samples early this year. For us, that typically kicks off a process of porting the code, adjusting functionality where required, and then a significant process of tuning, tweaking, testing, and iterating. Intel engineers consult and support us in this, which definitely speeds things along. It’s still a work in progress but we’re seeing some amazing early results.

First off, on the initial port of the code, we measured a 50% increase in our streaming packet processing engine – the core engine of the appliances. On further examination, the improvement was delivered mainly by two new capabilities:

Intel’s latest Sunny Cove M microarchitecture increased the number of instructions executed per clock cycle
It also increased per-core in-flight memory loads and stores, which has always been a focus point for us in our search for ever-increasing performance

It’s also worth noting that the above was achieved with little specific tuning of the code to utilize the new capabilities. That work is now ongoing, and we expect further significant improvements.

Another area we (and our customers) care about is compression. Compression to disk both increases the effective data rate that can be written to disk, plus increases the amount of historical data that can be stored locally for instant retrieval and reporting, per unit cost of course. With Ice Lake’s direct support for Bit Algebra and Vector Bit Manipulation instructions, we’re seeing almost 2x improvement in this part of the pipeline, and looking to get to 100Gbps sustained. More to follow on this.

And perhaps the area we are most excited about is Intel’s new Speed Select capabilities, including both Base Frequency and Turbo Frequency in 3rd Gen Intel Xeon Scalable processors Ice Lake. This allows the application to select certain critical cores to run at greater “Turbo” speeds than before, while still keeping the whole unit below the power/heat envelope. This is so important because as you look to performance tune a complex, multi-threaded, high-streamed real-time application, you always end up with threads that run hotter and are the bottleneck to overall performance. Even after you have rebalanced all you can, this will still be the case. So then you eventually have to rearchitect (which is really expensive in time, and you still end up with hot threads!!). With this capability, you can ensure the critical threads always run at maximum speed.

We’re looking at this initially for our timestamp processing. Timestamp processing is computationally only moderately expensive, but when you want to process 100 million per second, that’s a lot! And there are massive benefits to doing this sequentially on a single thread – multithreading it introduces a lot of avoidable headaches. So with Select Speed we’re looking to achieve the magic 100 million per second, while avoiding the headache of multithreading.

A lot more to come here. As we get further into the details, we expect to uncover further benefits and gains, but our first impression of 3rd Gen Intel Xeon Scalable processors Ice Lake with Corvil Analytics is exciting and impressive.