How to Run 20-plus Pre-Trade Risk Checks in Under a Microsecond
Tue, 20 Sep 2011 09:02:00 GMT
An Interview with Matt Dangerfield
With the recent introduction of SEC rule 15c3-5 (which requires all firms who provide sponsored access or DMA to ensure their clients’ orders go through adequate risk controls), vendors and service providers have been bringing some increasingly innovative solutions to market to help firms comply. However, forcing orders to go through pre-trade risk controls generally adds latency. So how can that latency be kept to an absolute minimum?
Front office managed services provider Fixnetix has recently been demonstrating its own solution to that challenge. With a latency of just 740 nanoseconds for running 20+ pre-trade risk checks, the firm believes its iX-eCute chip is now the world’s fastest trading appliance.
In this interview, Mike O’Hara speaks to Matt Dangerfield, Director of Trading Solutions at Fixnetix, to find out more.
HFT Review: Matt, you claim to be offering “nanosecond trading” through youriX-eCute chip; your recent press release mentions 740 nanoseconds wire to wire. What exactly does that mean? What are we measuring here?
Matt Dangerfield: We’re measuring the first byte coming in from the order to the last byte going out to the exchange, that’s the latency through iX-eCute, including all the manipulation, the look-ups, the pre-trade risk checks and session management and everything else.
HFTR: What kind of pre-trade risk checks are we talking about?
MD: A lot of the risk checks are standard ones. Europe and Asia have been using them for quite a while and are they’re now being implemented in US. We’ve spoken at length with both the SEC in US and the FSA in UK, as well as other regulatory bodies, to show them what we’ve got just as an FYI to engage their response, which has been very positive. Although the regulators haven’t always known exactly what they want, if you look at the recent FPL (FIX Protocol Ltd) report on pre-trade risk best practices, all the risk checks that are in that report are in our product, as well as some extra ones. So for example we have quite complex ADV checking and profiling to the exchange.
HFTR: How are the risk profiles actually defined?
MD: We run several risk models in the sense of pre-risk checks on the client order flow coming in within our client manager, and then we have an exchange profile manager which does risk checks too. For instance, the LSE would have a rule whereby you can’t submit an order in the open that’s five percent away from the previous close. And during continuous trading, you can’t be more than 10 percent away from the last market movement of that stock that you’re actually trading. So therefore, we enforce that on the exchange venue manager profile as opposed to the client profile, in case you’ve got multiple market orders coming down that one logical session.
Some clients have dedicated connections, so they’ll have “FIX Session One” or “Native Session One” for one market and then you can profile it in the client manager. But if they’re sending down orders for EuroNext, NYSE, LSE, Chi-X, BATS all down one connection, you don’t want to interrogate every order that comes down for the market profile risk checks. You need the market gateway to profile those checks so you go through the standard client risk checks or pre-risk checks that you want to do and then you do the market-specific ones on the market side. That way, you don’t slow down anything that’s happening in the client manager side.
HFTR: In terms of the 740 nanosecond “wire-to-wire” latency, how do you accurately measure that? And how can clients be sure that’s what they’re going to see too?
MD: There are several different ways of measuring latency. There’s optically tapping the lines at either end of the box and then bringing it up to a third party, or TCP/IP dumping and then looking at it Wireshark or something like that, or using network port replication or mirroring to get the reports off to another area and then analyzing it afterwards. Some of these methods add overhead to what you do, but if you know what that overhead is, you know what the delta is, so you move it away and then you’re okay.
The 740 nanoseconds wire to wire was actually captured by independent software that we actually have quite a large deployment of in Europe. We confirmed it with doing TCP/IP dumps and then we used Chipscope to measure exactly from the first byte in to the last byte out to the exchange.
Regarding the second part of your question, we are completely honest and truthful about everything that we do in the latency world. So what we quote, we allow people to come in and see. And when we deploy or design a model, we actually write these numbers into an SLA so we’ll get financially penalized if we don’t meet those SLAs.
It’s very, very important to us that when we go into contract with a client, there is a testing window where they test not only our execution capability through our software or hardware but also running on our infrastructure and our line, because a lot of people think of this as just a device. But there are lots of other things you need with it. You need the fast market data. You need the infrastructure we have. You need the location we’ve got. All of these add to the whole experience.
HFTR: So how would your clients typically be implementing this?
MD: It depends on the client. If it’s a prime broker model and it’s sponsored and they’re using somebody else’s ID, an MPID of membership, then it comes down to the question of what that prime broker will allow the client to do. 9 times out of 10 they won’t let them go anywhere near the risk checks for security reasons. They don’t want them to control it, so it’s all locked down, the prime broker takes care of it and the clients just connect in and send the order flow.
If it’s a proprietary or a member firm that wants to do this under their own stream, generally they have their risk checks further upstream in their own CEP engines. So they only want us to do a “catcher’s mitt” for the wild ones that occasionally they miss – or could potentially miss – so we’ll apply them and then they can control the different tolerances. For example, if they wanted to trade warrants on the SWX, then rather than having a 10 percent away from continuous trade or last trade, it would be 25 or 30 percent because of the swing of the warrants percentage.
HFTR: How wide-ranging is the instrument coverage on iX-eCute?
MD: We don’t really care, if that makes sense. The reason we don’t care is because it’s just a protocol, although the actual risk checks are different between equities, futures and options, FX and so on. It’s a case of whether we do contracts or notional values or fat-finger checks, whatever those checks are, we will just profile it to the exchange. What instrument the firm is actually sending through, we don’t really look at. We’ll just make a subscription for the market data and then pass it down, so we have the market data and the movement in there to be able to check against. In effect, it’s just comparing one number against another.
The reason we concentrate mainly on equities at the moment in the US, is because of the regulatory requirements that have come out with SEC rule 15c3-5. We sell a service and this is all about service. We offer the service of pre-risk, fast processing to our US-based customers or anybody who wants to execute in the US because they have more of a pressing need. Europe has a pressing need for the speed and risk restrictions too, but MiFID II hasn’t come out yet. I think there’s a consultation paper due out before the summer recess, which will open up and specifically address some of the HFT discussion points. Once that paper is released we’ll probably see more activity in Europe for the pre-risk side. For the connectivity side and the speed, obviously Europeans and the Asia-Pac countries want it but we are concentrating mostly on the US cash markets for now.
HFTR: Can you tell us a little bit about the technology behind the chip? I read somewhere that you’re operating four times faster than standard field programmable gate arrays (FPGAs) but I understand the iX-eCute is actually an FPGA-based solution. So how does that work?
MD: The way it works is we use the Xilinx Virtex-6 chip, which is Xilinx’s latest chip; they’re working on Virtex-7 now which will be out later this year. We basically take one of these chips on a customized board and it’s the customization of the board with the different memory blocks on there and the layout of that board that gives us the acceleration that we look for. We didn’t actually design the PCB ourselves, although we do have a company on retainer that would do that for us so if somebody wanted us to design a board from the silicon up and sort all the roots out, the endpoints, the gates and so on, so we could do that. But generally, we’ll go to a mass manufacturer and tell them we want this chip, this type of memory, we want it in this type of area etc, and they create that design for us. Generally within 16 weeks they can spin a card off, depending on the complexity of the card.
Then the programming of the card, whether in VHDL or Verilog, is all down to us so we’ll put our own patent pending IP onto the card. The way we utilize memory and gates to give us the speed is our “secret sauce”, so to speak.
We have a module base built for iX-eCute, the client manager, the market data manager and the venue manager. There are other different bits, we have a north part of the card, a south part of the card, different protocol engines and different components in between, but in general we sum those out as being in three parts: the venue side; the client side; and the market data side. Then we build up the profiles and the protocols inside. For instance, if you want a FIX engine we’ll handle the whole of the data dictionary 4.2 and then we’ll put customized tag ranges in there (e.g. 7000, 9000, 10000,) and then add in some extra space to cater for future growth. Then if you use that space up, we’ll hardcode that across into the data dictionary properly to free up the flexibility of the spare tags so you can do more manipulation if and when you want to.
HFTR: What other kind of customization is available and how long might it typically take?
MD: It depends on the customization you want to do. When we talk about customization, if you want a tag enrichment suppression, then we can do that in a matter of hours. If you want a whole new data dictionary added in to facilitate a load of tags that you need to cache for your middle office system on a drop copy, then that can take hours and if you want a complete protocol engine rewrite, then that can take days or dependent on the complexity of it, we might take that offline and do it in a different way. In fact, by end of Q1 2012, customers will be able to use iX-Eye to manipulate their own environment.
HFTR: What is iX-Eye exactly?
MD: iX-Eye is our command and control GUI which is built using push technology to facilitate the updates. In essence, it’s a magnifying glass that allows you to tap in to any execution management system you want. We can bring all your order flow – whether it’s in native format to an exchange or whether it’s in FIX – and display it through a single front end. iX-Eye has its own market data so it does a mark-to-market for risk rules and profiles with alerts and pop-ups. So whatever happens in the system downstream basically updates the front end allowing you to step in, do single order cancels, mass order cancels, restricted list, adjust intraday limits and so on, all in real time. You can extract information out of it to do reconciliations, you can drill down into the FIX message right down to the FIX tag or if it’s in binary form, right down to the binary if you know what you’re looking for.
And it has complete audit tracking capability. We also can tap it into sponsored access interfaces as long as we have permission from the exchanges and the MTFs that granted us permission to tap into their APIs directly without going through their front-ends or their Web interfaces. So, when a client is trading sponsored access through your membership and they’re doing DMA and DSA through different systems, when you say “cancel trades” for that client completely, all your sponsored access, your DSA, your DMA gets shut down simultaneously by one GUI, so you only log in once, do it once and that’s all audit tracked and password authenticated. We profile that as well, so a risk officer or a compliance officer will have a different view to a support person, trader or sales trader for example.
One important thing to point out is that we don’t corrupt messages to the exchange. We don’t say, “Oh, you broke a risk check, I’ll corrupt the message and let the exchange reject it”. You have a full session management so if a limit or a restriction is breached; a formatted message goes back in the same format and the same protocols it was sent in. That’s very, very important, because some exchanges will not give the messages back in the right order, so we’ll reorganize messages so you only get the messages back in the correct order. If an order goes into the market and goes into a pending state, we’ll queue those messages until the exchange either acks it or rejects it and then we’ll pass all the other messages. Let’s say you sent an order in and it’s in a pending state because there’s been a problem on the exchange and suddenly you get a reject or a fill back, all your preceeding messages will have properly formatted reject messages against them, so you don’t have to do anything on your engine. Your engine will understand it and will be able to process it while still maintaining it’s session integrity.
HFTR: Looking ahead, what are your plans with regard to new markets, particularly around iX-eCute?
MD: It’s quite interesting in that iX-eCute can hit any market that supports FIX, and within six to eight weeks for a market that doesn’t support FIX, we can write you a native interface. That means FIX in, native out or native in, native out. We’ll change protocol engines to handle whichever one you would like to do. So if we don’t have a particular market yet when you contact us, then it’s only six to eight weeks to do it.
We are very much customer-led, so if you want an Asian market, we’ll code that market for you, but we’re not coding every market just for the sake of coding a market, the demand has to be there. And markets change. Like for example with BATS and Chi-X, if the merger continues, they’ll be coming out with their own protocol. So, is it better to do that protocol now or to use FIX? The speed we get through doing FIX to FIX through our product gives you that advantage almost like native anyway. The only time you see a major uplift is if you go native through our product as well.
But when we talk about speed, this isn’t just about HFT. If you can improve the speed of general DMA flow and DSA flow, not all DSA needs low latency, but in general if you want to execute something, you want to do it fairly quickly with minimal impact against market performance and if you can do that in small clips under the radar or just away, then this is the type of product you need. From the feedback we’ve received so far, especially from SIFMA, which went exceptionally well for us, we seem to be at least 6 to 12 months ahead of most of our competitors.
HFTR: In what way?
MD: First of all, because iX-eCute is absolutely real, you can hold it in your hand and this one chip can handle a million messages a second. We like to talk here about “determinism”, regarding the performance of the chip whether it’s one order or a million orders going through. If we put 10,000 or 50,000 orders through and it takes 740 nanoseconds, if you go up to a million, it’s going to be like 740.11112! So it’s very, very granular in that sense. That’s what hardware does.
Don’t get me wrong. It’s not the easiest thing to do. We have worked very hard on this. We have an absolutely outstanding and excellent development team here that understand it and the engineers that we have working on it are focused on solving engineering problems and this is an engineering problem, getting something from point A to point B and back again, not so much a financial markets problem. These guys have been doing this for years in telecommunications, in geo-orbitary satellites, in telemetry systems, in target acquisition systems. So we’re just taking the engineering concepts and applying it to the financial markets.
HFTR: Finally, what are the options for firms who are interested in using your services, particularly around iX-eCute?
MD: We offer iX-eCute as a fully integrated product with everything else we have. So it really is an OEM solution. It can be purchased as a standalone product but you get the power from it and the security of what it does and everything else by taking all the other services. We’re a company that sells services, we just happen to have quite a large product suite and we wrap the service around each one of our product lines, so if you have your own EMS system or whatever, we’re not trying to take over the world in that sense. We’re just trying to help our customers improve their performance by letting them choose what they need. If they want to take the whole suite, they can take the whole suite or they can just cherry pick what they want and we’ll wrap it up for them.
This isn’t myth, this is real. We’ve got two centers here and in New York that people can plug into and test this. The proof of the pudding is in the eating, and that’s why we always do a trial for customers before they buy. If you don’t believe us, come and connect. Come into the offices here. Once you’re happy with that, we’ll get you some form of connectivity and we’ll deliver your logs, so you can see everything in real time and run your own scripts against it. You can run your own order flow against the market and see what it does; you can see your prints. We’ll give you aniX-eye too, so you can actually play with the command and control yourself from your own desktop without any interference from us. And then when you’re happy with what we’ve actually done in a technical implementation, we’ll tailor it to your needs.
And that’s very important because for people to be able to actually test things out for themselves and to see it working is very powerful.
HFTR: Thank you Matt.
For more information on Fixnetix (ranked by the Sunday Times Tech Track 100 as the the 1st and leading “pure” financial services technology company for trading, data and risk management in the UK) , please visithttp:/
This article was sponsored by Fixnetix.