Speed of light is approx 1 mic for 1000 feet. There's more than 1000 feet of cable between your server and the exchange. So I'm talking about internal latency. I don't think 1 mic is achievable in software. 3-5 is barely. As long as there's no thinking involved.
mmt - since you mentioned you colocate with your FCM I suppose your executions route via the FCM's TT setup (ie cross-connect to FCM side TT price and feed servers) and you dont have DMA? I am just curious to know what kind of latencies you are seeing there as we have tried similar setups with TT and another provider and if your latencies are similar to ours (ie NOWHERE near DMA executions) then internal latency will be the least of your problems! For us it was a conscious step to move away from DMA as it didnt matter (as much) for the current batch of strategies - ie decent latency is enough but to be honest considering what these guys charge and our overall spend for tech the execution latency is extremely poor relative to what we used to see. Note when I refer to execution latency I mean full roundtrip from Client Req to Exchange Ack.
We are not using TT any longer for reasons unrelated to latency and more to do with access to markets. I think you do not need the TT FIX Adapter for DMA. At the moment we are with CQG using FIX Connect (order routing via FCM also) for executions and I find it pretty slow (again relatively speaking). To be fair though, I have not done a proper investigation into this as so many other things to do but skimming through our logs and taking mental notes of waypoint timestamps my initial impression is not positive. I don't know TT's Autospreader as we are API users so I dont know how easily available execution logs/ details are. Have you got an idea what your roundtrip time is?
mmt - that would confirm my observations with TT from last year. I'd see latencies in the 10-20milliseconds range. Note this is not really TT's fault as I know that TT in a DMA setup has pretty good latencies. This is not from personal experience with it but from talking to another trader who uses TT for EUREX DMA. I suppose sharing the FCM's TT servers, hosting infrastructure, network and connectivity with other traders simply comes at that cost.
I guess I don't understand the question. If you are getting picked off during fast markets and are otherwise doing fine, why not turn off the autospreader during fast markets? "Fast Markets" are probably news related which can be somewhat anticipated. Sure, fast markets is where all the money is, but the amount of money and skill to needed to compete here are daunting. If you want to be in the space where you can trade fast markets, then you are asking an impossible question. You can't even come close to competing in the HFT space with software like TT. When you are dealing with being 1-microsecond faster than someone else (or even more ridiculous), even the wrong settings on your network card can be catastrophic. And the list is long of things that can add nanoseconds to your latency. If you don't have this level of skill, or can't hire it, you migth as well try to get to Mars because that is probably easier. I have the knowledge to compete in this space, and I can tell you that the attention to detail is bordering for example on being able to take an oscilloscope to your switch and measuring latency. If that is alien to you, seriously, you are risking your financial well being by thinking that TT and "closer to the server" will make any difference. In this case, I agree with the person above that said try to be more clever than the next guy, don't try to outrun him.
Previously I worked for a options market making firm. I am profitable trader. I use off the shelf solutions right now, but this does not mean I should only concentrate on making smarter strategies. I can decide for my self, whether the costs will outweigh the benefits. But, since I am new to this field , I need some guidance.