Biggest bug

[crimsonite]

I've had huge problems trying to save optimization and forward walks... for large opto's, it takes several minutes to save a test, and sometime sit does not save at all. For whatever reason, it uses incredible resources to save, especially disk drive resources.

It also takes a few moments to load backtest data and opto data off of the strategy analyzer.

I hope this is fixed sometime down the road, but it does give the impression that parts of the strategy analyzer is spaghetti code.

[NinjaTrader_AdamP]

Crimsonite,

How large of optimizations are you running? I just did an optimization on the EURUSD with a strategy that did 250 back tests in total. When I save it the directory size changed by kilobytes. What version of NinjaTrader are you running?

I look forward to helping you resolve your issue.

[crimsonite]

Most of my optos are on the six default contracts for 12 month periods. I set them to run for 2-4 hours in most cases, but sometimes longer. I ran a 14 hour opto overnight earlier this week, and am currently running a 60 hour opto that's scheduled to complete Tuesday early morning. Total backtests are generally 50,000-1,000,000.

The length of the opto generally doesn't have much to do with how long it takes to load.

IMO, these are most certainly not "too large," since guys running Bloomberg are connecting to their 4 terraflop floor server running million+ point optos in a few minutes flat.

To be clear, I'm extraordinary happy with Ninja and believe it's the best retail platform for traders executing their goals. I left a positive review at elitetrader a couple days ago.

[NinjaTrader_AdamP]

crimsonite,

A Core i5 @ 4GHz is something like : 10 GFlops, and 4 terraflops is 4000 GFlops. This is quite a big difference since it may be a cluster that has more processors, higher throughput as well as larger RAM capacity. It doesn't scale necessarily linearly due to the differences.

Are you mainly having issues with the speed of optimizations or the fact that its taking up too much room to save them?

[crimsonite]

To put this in perspective, a single Tesla card delivers 78 Gigaflops, and a montherboard can take up to six of them. An Intel 950x does over 100gflops - so around 680 gflops for a single motherboard.

Bloomberg terminals work as a client-server, not as a workstation. They usually connect to a Oracle server in the floor head end room.

A firm at 23 Wall St runs 128 terraflops, just for fixed income.

I really like Ninja, and I'm grateful the initiative was taken to develop it for retailers (that's why I'm so quick to point out bugs ).

I'll try to post a screen recording the next time I save an opto. I hope I can help along the way. Best of luck, C

[NinjaTrader_AdamP]

Crimsonite,

Yes, I am familiar with some cluster setups from my university days. I used to work for a biomedicine lab as a research assistant where we would run projects that would take days sometimes. We had 4 different clusters at our disposal but we shared some of them with other departments. I ran some individual projects that took several days before but I was limited to 8 cores on an Opteron cluster since I was given lower priority and system resource cap over others.

Anyway, if you find anything related to your optimization issue, I would be happy to help you resolve it.

Please let me know if I may assist further