Futures trading latency

cjbuckley4

cjbuckley4

to anyone here who uses a vps or dedicated server in Chicago, I'd like to know the following information:

1. What firm do you trade through?
2. What data provider do you use?
3. What latency do you experience receiving data?
3. What API solution do you use (i.e. TT/TT FIX, CQG etc).
4. What latency do you experience placing trades at Chicago exchanges (and CBOE if you have that info).

I'm not trying to compete with firms for who have Aurora colocation and network cards worth more than my account balance. I'm simply looking to understand what conservative estimates I can make about latency in order to make the most accurate assumptions possible for Backtesting. I initially thought the best way to get this information would be to call up brokerage firms and ask them, but after contacting 3 and having some conflicting/ridiculous answers as well as being confused with a 'serious' Aurora colo customer a few times, I realized this would be a difficult question for them to answer. Again, to be clear, if you could exclude the wire-to-wire latency of your particular system and simply discuss the time it takes to receive data and the time it takes to execute trades, I would appreciate it, but really any ballpark figures would be appreciated.

Also, if any Aurora colo customers stumble on this, would you mind helping me understand the cost of colo there and trading at CME? It's just a curiosity, but I've read all sorts of stuff. To reiterate, I have no business calling up CME and asking them, but I'm still curious.

Thanks
 
Such as steadfast? That's the type of service I would consider using, but wouldn't they only know about what their ping times are to certain exchanges and maybe to some data providers? I'm not concerned about what I understand to be a <1 millisecond ping time. I'm concerned with, for instance, an anecdote saying that TT with Advantage Futures has at least 60 milliseconds latency. Maybe there are some other limitation even higher than that that I'm not aware of, so I wanted to ask the community here to make sure I was making accurate assumptions before I jumped in. So far, it appears that I'll probably be using TT with Advantage Futures, but I wanted to get an idea of what other people are using as well. To be clear, again, (for folks who will inevitably come later and warn me of the dangers of competing with HFT participants) I'm not necessarily saying that I'll be targeting opportunities that require 60 millisecond latency, I'm just trying to understand what assumptions are reasonable.
 
I know you are looking more so for the speed through brokers, but just in case you want to see I believe the CME has the speed at Aurora on their website, not 100% positive on that though. I went to a thing at their headquarters and one of the presenters talked about their specific latency and latency through a few other means. It was extremely fast to say the least. If I can find the stuff at home tonight I will post it.
 
Thanks for the info @FCXoptions. I have seen some information on CME's matching speed, but not on their website. I'll look. As you say though, what goes on inside the Aurora datacenter and on CME's matching engine can't really be that much of a concern for me, because my broker will probably be talking milliseconds or seconds where they'll be sweating microseconds or less. One of my reasons for becoming interested in futures trading was that I loved CMEs transparency though. Maybe it's naïve of me to assume this, but they seem to give way more information about matching, colo, etc than other exchanges.
 
Some FCMs offer collocation or equipment rental directly. Such a setup is likely to provide you with the lowest and most consistent latency to their systems - but of course this should be verified and monitored to see that it remains so.

Using a third party equipment and hosting provider for a VPS or a dedicated server generally means that you will also make use of their internet connections between their equipment that you rent and the market data and trading system used by your FCM. Run a traceroute to learn what path your data travels and do this from time to time. We had a customer who rented a server from a hosting provider who was located on the same floor as our data center at 350 East Cermak in Chicago and he complained that market data from us was delayed and his orders were taking too long to get to the exchange. After verifying that the internal latency of market data and orders within our systems was normal and very low, we ran a traceroute from his machine to our systems. We found that the path his data travelled took his data from Chicago to New York and then back to Chicago, even though his machine and ours were in the same building and on the same floor. His hosting provider, who also provided him with internet access, used a number of ISPs and one of them at that time routed all traffic going to Chicago through New York, regardless of the origination of the traffic.

We have also seen traders who rent a VPS or a dedicated server from a hosting provider in Chicago have long transit times to our systems running in Aurora. We even had one case where a trader rented a VPS from a hosting provider in 350 East Cermak in Chicago and he also rented a VPS from us in our (the CME's) Aurora data center 30 miles west of Chicago. The trader used the Zen-Fire system at that time and found that his orders reached Zen-Fire, which was located at 350 East Cermak, much sooner (tens of milliseconds) when they were submitted from our VPS in Aurora than when he submitted them from his Cermak VPS (even taking into account that he would trigger sooner in Aurora).

The long and the short of it is that when using a hosting provider do not assume that physical proximity provides you with lower latency. And do not assume that the path travelled between the machine that you use and your FCM's system remains constant. Ask the hosting provider for evidence of time between his machines and your FCM's systems, verify it and monitor it from time to time to make sure it remains so.
 
That's a really good point @jjw . A host like steadfast was only one of the things I'm considering. I've seen Advantage Futures colocation information and pricing for Aurora and it's actually pretty attractive. That's probably going to be my number one choice if my trading necessitates low enough latency to justify the additional cost. That's still a long long way away though, I'm just feeling out the possibilities. I was surprised by how affordable their offering is though, because I previously read CMEs colocation literature and it looks like that monthly bill would be in the 5-6 figure range depending on your needs. Advantage apparently also has hosting options as well which I might look into for the reasons you mentioned. Given my account balance, their hosting may well be the most realistic option.

To be clear, I don't even think hosting will really impact latency in a meaningful way, the biggest pro of hosting is probably just uptime and redundancy. Colocation is what will really decrease latency, but again, I'm not sure if that's gonna be necessary. If you want to do something truly low latency, colocation is probably the easiest part of it. Network cards, customer drivers, linux kernel mods...that's not a 1 man job and it's above my abilities. I'm not sure you can even access CME without a front end like TT, CQG, etc with under 500K, but don't quote me on that. I'm just trying to feel out what's doable at the retail level without breaking the bank so I can make reasonable backtesting assumptions.
 
Last edited:
With respect to "access[ing the ] CME without a front end like TT, CQG, etc with under 500K" most commercial systems, even those used by single traders and small houses, tend to provide libraries (an api)that can be incorporated into a trader's proprietary software. Combing that with collocation is a relatively inexpensive way to get "up close and personal" with the exchange and its data. We have many customers who rent our VPS and dedicated servers precisely so that they can run their own software, connected to our infrastructure with less than a 1 millisecond hop.

Other traders seek even lower latency and I agree that network cards and kernels and machine tuning then becomes necessary. In such cases we provide assistance and recommendations on how to organize code, tune machines and make use of kernel bypass mechanisms (and we can even provide appropriate hardware). We also have a tap so that latency can be measured independent of our switches and servers. And even this can be had for far less than $500k.

It gets really expensive when ultra ultra low latency (sub microsecond) is required. That's when FPGAs and other interesting setups are required.
 
You must log in or register to reply here.
Back
Top