To expand on this in the nicest possible way, let's talk about microseconds. There are 1,000,000 of them in a second. To do latency arb you're working in the millisecond range. You're doing things like buying dedicated microwave tower links between exchanges because the speed of light in fiber is slightly slower than through air. We're talking the speed of light here, just to be clear on time implications. On the hardware side you're talking FPGA hardware with software that's optimized for the parallelism FPGA gets you, again a regular CPU/GPU isn't fast enough. So if you're asking about ssd vs traditional hard drives in the same sentence with latency arb, with all due respect you're in so far over your head you don't even realize how little you know (the OP, not algofy).
From other posts I presume you think you can work on the second scale with "slow" forex bucket shops. From having tried this, I can tell you that anyone with an API has a fast enough feed that they're comparing your order to the current price when your order arrives and microseconds isn't fast enough when you take into account the delay for you to get the quote, process it, send it back, and them to receive it. If you're dealing with a bucket shop with stale quotes, even if you play two of them off each other you have to move money between them so often that they catch on within a few thousand dollars of profit. Plus you have to reverse engineer their data feed and build your own engine that mimics theirs as far as their server is concerned, which takes more time than it's worth considering they'll shut you down before you recoup the cost it took you to do the reverse engineering.
