Do you hint you were programming in assembly language as opposed to relying on an optimising compiler?
You have a good imagination... But no, I wasn't... But even if I was, it illustrates clearly the case that fattened, full of high-fructose-corn-syrup .NET jittered code is bad for your health What I said could mean other things too
Well, we all know that x87 coprocessor and all the follow-ups were operating internally in 72-bit floating point numbers. Any floating point operation between 64-bit or 32-bit numbers involved conversion into the 72-bit format and back. Naturally, any code that relied on loss fo precision when converting to 32-bit floating point numbers and back to 64-bit numbers must have suffered a penalty.
I didn't know that everybody here knows assembler... I think you are stretching a little bit Any way these were the fastest instructions at the moment for double operations. In addition to that the data was as double in memory. Any other pre-conversion of doubles to then try to use the weird SSEX instructions is slower than using directly FPU instructions. And by the way I think you information is not correct, StX registers were/are 80bits and doubles are 64bits. What ever the internal conversion here.. you are talking at "electrons level"..lol...
Hehe! We can discuss details of past battles forever. What I am trying to do is to get you to say "This is our unique selling point. No one does it. I don't think anyone else will have the capacity to do it any time soon. This unique feature has a potential to make you lots and lots of money." I appreciate it may take time to form a proper marketing speech. No rush. However, saying "we combine 6 pretty common features that no existing platform combines but there are competing platforms that offer any 4 of these 6 features at a fraction of the price" is not a very compelling marketing speech. What price range do you envision for your trading platform? A Dyson vacuum cleaner, a Ferrary, someting half-way?
I assume that he meant that he either realized there was a redundancy of calculation so that the floating points could be run outside of the loop, or as was the rage, worked with all integer market prices (# of ticks so 102 15/16 = 16*102+15). As far as compiled code vs JIT compilation, I don't really care. I tutored a private group on x86 assembly language many years back, and I based most of my lectures on "here is the C code, here is the resulting assembly, let's understand how they are the same." Since the students were capable programmers to start with, it went more productively than discussing J-cells and bus architectures. The reason that I bring this up is that I was appalled at what poor assembly code was actually coming out. As someone whose Cray research time allocation was 10-100x my paycheck, I spent a large amount of my time making my code almost incomprehensible in an effort to tune it to the particular architecture of the machine. Heck, I still run my loops backwards as a matter of habit "for(int i=count-1;i>=0;i--) {}" Machines are so fast, so simple, and so cheap nowadays that it isn't worth that much attention. When I do try to squeeze my code, the biggest payoffs tends to be string handling and disk IO. The fewest string allocations the better even if that means concatenating it all at the end. Being able to read and write whole files at a time has made life so simple and so fast. Having gobs of memory is great. All of this is irrelevant to the original post though. Prospective customers want to see results, and they don't care whether you need more or fewer servers to deliver their working trading system. The key features were Dual server/desktop operation: Respondents have assumed that this software will be licensed to trading houses and run on their servers. Big assumption. If not, there are a myriad of problems such as "is my system no longer making money because someone else's system is hogging the bandwidth." If it is licensed, then it remains to be see if it is actually better than what they already have. This seems to imply that since your servers are 64bit Windows? that all of the desktop system developers must also use 64bit Windows. Not completely unreasonable, but some shops may need some 16-bit legacy stuff like TWAIN. Gotta remember that selling software is different from making software. Avoid giving the customer a reason to say no. Nothing new there. Even with systems like NANEX which guarantee to deliver byte-for-byte the same result (except some communication status codes) regardless. There are still communication issues, software upgrade conflicts, server failures, and general timing issues which can not be guaranteed. (If your particular data connection stopped properly receiving Level II quotes for an hour one day, would you really make people back-test without Level II quotes?) Maybe it is a good idea to come up with a stress-test battery where people can test their system against a script of anomolous complications, but guaranteed identical? This has been clarified to mean that you are providing support utilities to help the system writer judiciously accelerate their back-testing. Fine, but we get back to 90% of traders being non-programmers and 99% being alot more focused on their trading than their coding. Sellable value? I like the way that NinjaTrader has made caching transparent. May not be optimal, but it doesn't get in the way either. Not really sure what that means. There have been massively parallel Internet based cryptographic challenges that made sense, but when it comes to tick-by-tick back-testing, bandwidth costs. And if this is a shared server environment, rather than a licensed package, I see major problems with resource sharing. One trader selects a very specific strategy to run over a small dataset, and another brute-force cherry picks from an astronomical number of parameters over a ill-considered dataset like "All Stocks". Part of cloud computing is managing customer utilization to maintain the illusion of free, unlimited resources. Good Luck to you anyhow.
It looks like you love the phrase: "if you can't convince them... confuse them"... Don't tell me you don't like that phrase!! I really appreciate that you keep the exchange...And this is not personal... But I want to point out about the quality of your posts. They are like the beef loin flap at BJ's... red in the outside... with the fat hidden in the inside... I don't think you do it in a mean-spirited way, but in some occasions your posts have false statements "inside". This only degrades the discussion. My only point is: tell the truth to the people.. I see some sophism here.. half-truth mixed with half-non-truth to convey certain message. The purpose of my only thread here in this forum is collect opinions and comments from traders and developers. I have found a lot of "mythology" taken as historical events ... But in general I have found good information here. About your other questions, we are not open to discuss price, or monetization model at the moment because we simple don't know the best answer for that. And honestly this is not what is keeping us awake at night now.
Is it a more subtle way to say "divide and conquer"? Then it sounds familiar. To put your marketing pitch in the framework of a comic strip, you inspire a few laughs by connotation but the picture lacks a killer punchline.
These are all great points. And you have mentioned several problems that have cost the hair to more than a score of engineer and traders. And while my hair is still there I can tell you that these are the problems that are keeping me awake.