LOL, so help me, why would you then stream the data through sockets if on the same machine? Cannot use shared memory? Oh, right, synchronization issues, server mapping,...did I forget any potential issues? So yes, sockets may be simpler than sharing memory BUT you still ignore the big picture, you basically introduce a multiple additional potential points of failure and basically are forced to "hope for the best". And yes, I am probably more proficient in this space than you ever will be. I just think you make your life way too difficult.
Windows and Linux use different memory adressing schemes I.e the pointer values for the same block of memory would be different, thus ( to the best of my knowledge) ruling out shared memory. Sockets are plenty fast.
So yes, sockets may be simpler than sharing memory BUT you still ignore the big picture, you basically introduce a multiple additional potential points of failure and basically are forced to "hope for the best". And yes, I am probably more proficient in this space than you ever will be. I just think you make your life way too difficult. There is a reason no professional hedge fund nor ibank is doing what you try to do. Still good luck, I mean it, I have no reason to wish you anything but success.
As occam stated in a previous post, there are more exchange related failures than there are from running wine. As for you being more pofecient in this space... that is a bold claim considering you have no idea who I am or what I do, but believe what you want to. Im not here to measure up against anyone. If I had the capital that the iBanks do, I would purchase a direct exchange feed and co-locate(and run pure linux...). But I dont, I am a retail trader with a particular skill set that lends itself well to automated trading, a day job, and a significantly smaller bankroll. I would venture a guess and say some small firms are using a setup very similar to mine.
you mean you have proficient skills in coding, you do not sound like you really manage risk or have actually ever coded up alpha generating algorithms. Again, as I said good luck, in a low latency environment that you are seeking to operate in, such many points of failure are gonna be killer. Let that be told by someone who has worked in this space for many years...think whatever you want, I just tried to warn you.
How can you trade the book if you are executing through IB ? I've tested their gateway their fastest execution confirmation is about 200 - 300 ms. If you are lifting bids and hitting offers on CME at least i think it needs to be a lot faster than that right ?
I think if its on the same OS same machine, the sockets are treated differently. I can't remember if its linux or windows, but local sockets connections are not really tcp ip socket. maybe they changed that. Anyway using shared memory with all the pointer arithmetic is a bit eye straining. But yeah if your looking for sub 1 ms latency i guess thats one of the best ways. Correct me if I'm wrong but a socket stream will still have a minimum about 1 ms latency right ? With direct pointers its probably in the low micro seconds on a standard windows server. What ethernet card are you using for your low latency strat ?
Im using the book to create trading signals based upon Machine Learning (thats as deep as ill go), it is not low latency strat, but I will attempt to trade as quickly as possible given my latency considerations. Im not close to going live yet, but im using market orders so rapid confirmation is not a huge concern of mine. That being said, I would love to erase 2-300 ms from the conf time. Can you recommend another low cost futures broker with a c++ api? Im certainly not attatched to IB. As for sockets being treated differently on the same machine, you are correct. They are "local sockets". Windows calls them direct pipes, linux calls them something similar. What im not sure of is if I can use local sockets to talk to a wine program from a native linux program. Fortunately QT has classes for both local and tcp sockets that have very similar syntax, so when I get to that stage I will try both. I have no clue what my network card is. When Im up and running, I will run on a remote server, but I havent defined what I require of that server yet. Once my software is finished, back testing and execution modeling will help me to determine what my true latency needs are.