MT4 in a virtual machine ? - page 3

 
jjc:
There are multiple bottlenecks, and I'd expect an improvement in disk I/O to yield some benefit.

The cheap alternative would be to take some of your RAM and allocate it to a RAM disk. I'm sure I've done tests of this before and that it's yielded modest but non-trivial gains. My current machine has an SSD, and it slightly outperforms a RAM disk with Windows file compression turned on (which surprises me slightly). On that basis, I'd expect a RAM disk to offer a decent improvement in backtesting speed over a magnetic drive. The trouble, of course, is allocating a RAM disk large enough to hold an FXT file, even with compression turned on.

Well Raptor has a pretty serious machine and it is "only" 4x faster than mine. I can live with that for now, so I get a bigger upgrade when I finally switch machines :-)

I had read about RAM disks earlier today but even with 4GB on XP you only get 3GB you can use and I think you need a lot more than that to keep the hard-drive calm. I was thinking more like 20GB on a Windows 7 machine (to get the 64 bit address space). 10-20% speed increase doesn't do it for me :-)

So ... what's your hardware like on the speed test? Can yours kick Raptor's ramBUTT?

 
dabbler:

Well Raptor has a pretty serious machine and it is "only" 4x faster than mine. I can live with that for now, so I get a bigger upgrade when I finally switch machines :-)

I had read about RAM disks earlier today but even with 4GB on XP you only get 3GB you can use and I think you need a lot more than that to keep the hard-drive calm. I was thinking more like 20GB on a Windows 7 machine (to get the 64 bit address space). 10-20% speed increase doesn't do it for me :-)

So ... what's your hardware like on the speed test? Can yours kick Raptor's ramBUTT?

My laptop has a relatively puny i5-2467M processor, and was doing your test at the same speed as RaptorUK: 30 seconds. On that basis - and other experience - I'm agreeing that there are bottlenecks which increased processor power and disk I/O aren't going to doing anything other than chip away at.

Using a RAM disk instead of an SSD increased testing time to about 35 seconds. I managed to do your test using 1 GB RAM disk, with compression, but you'd struggle to do much over 1 year of data without more RAM to throw at it.
 
For most people the best gains are to be made working on their code . . . IMO . . . probably me included. The performance I get is acceptable at the moment . . . if it gets too slow I'll look at my code again ;-)
 
jjc:
My laptop has a relatively puny i5-2467M processor, and was doing your test at the same speed as RaptorUK: 30 seconds. On that basis - and other experience - I'm agreeing that there are bottlenecks which increased processor power and disk I/O aren't going to doing anything other than chip away at.

That is very impressive for a laptop. It's hard to understand how a laptop i5 can match a workstation i7. Thanks for the data.
 
RaptorUK:

When I get the opportunity I'll run the same test on my laptop, it is an i7 2640M, 2.8 GHz but will overclock itself to 3.4GHz it's only 2 core (4 pseudo core) . . . no SSD just one hard drive. If any of what I have said above is correct it should be similar, if not a bit quicker, than my desktop.

OK, as promised . . . i7 2640M, 2.8 GHz . . self overclocking to 3.4GHz . . using around 30% CPU, but that is probably an over estimation my CPU is running at around 10% even now and I'm not doing much.

12,442 trades, Profit factor 1.00 . . . Spread 0.0 ;-) took 17.6 secs start to finish.

 
RaptorUK:

OK, as promised . . . i7 2640M, 2.8 GHz . . self overclocking to 3.4GHz . . using around 30% CPU, but that is probably an over estimation my CPU is running at around 10% even now and I'm not doing much.

12,442 trades, Profit factor 1.00 . . . Spread 0.0 ;-) took 17.6 secs start to finish.

AWESOME!

 
dabbler:

AWESOME!


IMO CPU is almost always the bottleneck . . . just goes to show what 3 years difference can make in a CPU.
 

I'll do the test later tonight.

I'm so disapointed with the new Vertex3 SSDs... have moved my system back to HDD. Even running just the MT4 alone on a SSD caused nothing but trouble.

 
dabbler:

Let's try a speed test, and hopefully we can get some guys with SERIOUSLY POWERFUL machines to get interested and give it a go too.


Ok heres my result on your code, It took 1:20:

This was not quite as expected, It may be because my EA is out of the market most of the time and just keep running indicators and waiting for them to line up for entry.

I have plenty of RAM and a fixed size swapfile, and there is virtually no HDD activity during the test (after the fxt file is built).

This is what my 100% realistic EA produce, most cores are 'flatlined':

 
RaptorUK:

OK, as promised . . . i7 2640M, 2.8 GHz . . self overclocking to 3.4GHz . . using around 30% CPU, but that is probably an over estimation my CPU is running at around 10% even now and I'm not doing much.

12,442 trades, Profit factor 1.00 . . . Spread 0.0 ;-) took 17.6 secs start to finish.



Strange there is so much difference to mine i7 950 self overclocking to 3.3G.

Are you using Windows7 ?