Still the ballpark is the same. Not from the business side, but from the approach. You have to rip out anything that is standard because that everyone else can use - you have to beat what is easily available. HFT - be first, or be out. That is all I meant. Does not matter about the exchange because at the end that is what everyone there has. Fact is that this is a "be first" approach, and that means fighting because all the easy routes everyone can take.
Good point. Can't argue with that. Don't know much about some of the exchanges on OPs list, but I do know that the exchanges' latencies vary quite a bit and was hence interested from that perspective.
What I wonder is if garachen's group is only looking for 1 guy. The problem complexity is sufficiently high that one man might be able to optimize a specific use case, but still be overwhelmed with the complexity of when multiple exchanges are involved. The developer might have the expertise, but I think the cost of the project to be first virtually everywhere (particularly when factoring in the time to completion) is actually far beyond the cost of just one developer. That's disregarding the range garachen posted with his ad. That having been said, lower latency is always better, even if it is rolled out slowly for very specific use cases.
There is more than one extremely well known and heavily used exchange where just the trip through the matching is going to take 50x that number.
The basically the OP is asking for the impossible: kind of "swim the English channel and climb Mt Everest in the same day....." Correct ?
No. The processing time in the exchange is irrelevant when you do HFT. Point is - there is an event related to the exchange (x). You want to respond to that event BEFORE ANYONE ELSE. That the exchange then takes time Y to process that is not relevant to you - as long as you are first in the queue to be processed. This is the brutal thing. There is no "edge" because of statistical exploitation - the ONLY edge is speed, and this is an edge that really is about money. Faster computer, faster connect. And all the easy stuff (stock items to buy) are baseline - everyone already has that.
Your latency budget isn't specific enough, but if I guess what you're expecting correctly, this is quite achievable using a commodity stack. Don't see what's wrong with your proposed salary and why the other posters are giving you such crude responses - I wouldn't let them discourage you. If I was still in grad school, I'd probably have done it over the summer for a fraction of that.