An Interview with Chad Attlesey & Chris Butler (part 2)

Sponsored by Supermicro UK & Hardcore Computer

Last month, in part one of our interview with Chad Attlesey, President, CTO and Co-Founder of Hardcore Computer and Chris Butler, Business Development Manager at Supermicro® UK, we discussed how Hardcore’s over-clocked, submersion-cooled desktops and workstations are increasingly being adopted in the dealing rooms of HFT hedge funds and proprietary trading firms. In this concluding part of the interview, we talk to Chad and Chris about how and why their technology is generating interest in data centres and co-location facilities.

HFT Review: Chris, in the first part of this interview, Chad spoke of the challenges he originally faced as a start-up company trying to get into data centres, and the “Trojan Horse” approach he adopted as a result. What can you tell us about your go-to-market strategy with the data centres now?

Chris Butler: It is early days yet, so we are currently doing POCs with the banks, with a view to then deploying our liquid servers within the data centers.

There are a growing number of firms doing algo trading and high frequency trading who are co-locating their servers at the exchanges’ facilities. And we’re hearing similar stories from every one of them. Pretty much all of them have multiple sites around the world: in Mahwah, NJ; in Basildon and Slough, UK; in all the various exchanges in US and Europe. And so our goal is to go down an initial POC route, typically using the Detonator standalone box we discussed in our earlier conversation, to convince the firm that the Liquid Blade is a good vehicle for them to use in a data center with live market feeds. And the strategy is to then broaden that out on a worldwide basis.

We already have agreements from NYSE and Equinix, which are two of the data centers we have spoken with, to house the Liquid Blade as part of an ongoing deployment process.

HFTR: So if firms want to swap out their existing blades with this liquid-cooled technology, what exactly is involved and what are some of the challenges they might face?

CB: The beauty of the liquid server or Liquid Blade is that you’ve got two aircraft-style connectors on the back of it, the kettle-style power connector and your Infiniband connector (or whatever the I/O is). You can literally uncouple all of those self-sealing connectors in a matter of seconds, slide a unit out and put another one in there. So it’s an eminently serviceable solution for a discreet data center. The other point we made in our earlier conversation is that the actual hardware behind that is a very upgradeable platform for future generations of processor.

Chad Attlesey: It’s the same chassis; you can swap out Ivy Bridge-based systems in the future, so there is a direct upgrade path without a lot of expense. It is also using a standard Supermicro motherboard with single PCI slots so it can support a GPU or solid state disks should you want to use those as well.

We provide a cooling distribution unit (CDU) that is right in line within the 5U height that we occupy with four of our servers, two on each side with the CDU in the centre. That CDU is simply a high performance radiator fan and a smart cooling system, where we’re actually doing fan control and pump control monitoring. In the near future, we’ll also be providing remote monitoring capabilities for showing health information of those CDUs, so from any kind of co-lo environment to wherever you want to manage this solution, you can get that real-time information.

All the cooling is happening right there at the rack level because currently at the co-los, they’re not providing external plumbing and doing the various other things that would allow us to get maximum efficiency, but the main concern is the power/performance and that is what’s happening at the blade level. We can get the heat out much better, much more efficiently, and much more effectively with our cooling solutions, which will allow for the over-clocking capabilities and will allow for the additional power, in addition to that CDU.

And with HFT applications, we are using the high end Infiniband I/O capabilities along with timestamp capabilities built in. So it really is perfectly suited for the trading environment, it has all those requirements. We are optimizing I/O; we are optimizing core performance and storage access. Everything else is completely optimized for the best response time possible.

CB: A lot of money is spent around the world on lowering latency by the best fiber routes, by optimizing software and so on, but the blade technology we are offering is the missing link in increasing speed. If you are paying for a low latency network, you are obviously going to want a fast server on the back of it and the beauty of these solutions is that you are running your native C++ code on those. So there is no software adaptation required at all, you just run the same code but faster.

HFTR: If Liquid Blades take up less space, if they’re faster, they’re cooler, they’re quieter, they run with more predictable performance and they’re more reliable over the long term than exiting air-cooled units, it would seem like a no-brainer for data centres to put these machines in, especially if it’s just a case of swapping one box for another. But what about the cost? How much is the Liquid Blade technology in comparison with what firms might already be using in the data centre?

CB: The cost is in the order of about 20% more than a high end server solution from another manufacturer, so it is not prohibitively expensive. The extra cost associated with the metalwork extrusions etc., is effectively a one-off cost, because there is an upgrade path and we do include that as part of an SLA should the customer want it.

CA: But if you look at it for example on a CPU by CPU basis, we are actually on price parity. We’re largely a component driven cost. The cooling distribution unit has an associated cost with it, but then again so does your air conditioning and other things that go with the operation, so there is a slight offset there. If this were to be a plumbed solution in the data center, we’ve actually done TCO studies where the total cost of ownership over a three year life cycle is actually 30% higher for an air-cooled or hybrid-cooled solution versus ours. So actually you get much better return on investment with our product.

HFTR: So would you say this technology is “game changing” for the data center world?

CB: It certainly is, but what you’ve got to bear in mind is that the incumbents are extremely well established. They give a very high level of service; they’ve got good kit, etc. We are the new kid on the block and so we are currently playing our performance card as our route to market. As Chad said, with this self-contained 5U four blade solution and the heat exchanger in the one box, it is still actually dissipating the heat into the air conditioning environment of the data center. In an ideal world you would have oil to water intercoolers and that heat would be going outside.

CA: Or you’d have direct plumbing of the oil directly to an oil chiller or a geothermal generator heating your office space in winter time, preheating boiler water for the facility. There are a variety of other ways to recover and reuse that energy.

CB: Exactly, but we don’t want to run before we can walk. We are coming in and saying, “We can give you roughly 30% more performance than an air cooled system possibly could, and we can give you high reliability and longevity”. This is a very, very compelling argument, but today, it’s primarily performance-driven.

The longer term play is that in five years we want to be the ubiquitous cooling technology in data centers.

HFTR: Chris, you said yourself just now that all the incumbent suppliers are well-established. They have worldwide networks and substantial support infrastructures. What are your plans to support the growth you’re aiming for?

CB: Boston, our reseller, is a worldwide supplier, with offices in the UK, Germany, Paris, India and they are setting up in New York as well. So part and parcel of our SLAs is that Boston will provide 24/7 support.

CA: And if you look at the pedigree of the people that we’ve got working at Hardcore Computer, we have people who have run billion dollar contract manufacturing organizations. Our CEO and COO both come from that pedigree. And we know how to take things from napkin sketches to full-scale international production in a short amount of time, with very little resource requirement. We have the supply chain worked out and we are spending years developing this as a product, not just as a quick demonstration technology. This is actually a fully baked product that we are scaling and taking to the market.

HFTR: How many staff does Hardcore currently have?

CA: Roughly thirty employees right now, but our manufacturing capability is scalable and we can take and replicate our manufacturing process anywhere in the world where we need to go.

HFTR: Finally, what can you tell me about your patents?

CA: We have six US patents currently, other international patents and many more in process. We effectively own the submersion cooling market. We own the technology, so anybody else who is claiming to use submersion cooling is actually in violation of our patent rights. We take this very seriously. We have spent a lot of time, money and manpower to develop this into a product, we’ve done the engineering and research on it, we’ve done the testing, done the work around compatibility, all these things that are required to really provide a reliable viable product.

HFTR: Well, I wish you much success with it. Thank you Chad, Chris.

For more information, visit http://www.hardcorecomputer.com and  http://www.supermicro.co.uk/