Benchmark for NT

[ctrlbrk]

One year of 15 minute data should be no problem with regards to memory footprint, it will just take some time to optimize.

Another advantage of VMware is the ability to just pause the entire VM if the need arises, and resume later, including in the middle of a long optimization job. For instance on a multi-day optimization job, I sometimes want to free up CPU resources for other things here and there and can just pause the VM.

I am a big advocate of VMware. But I am not a typical user.

Mike

[KrisS]

VMWare is excellent from what I hear, though I have no experience with it. I use MS VirtualPC 2007. It is a free alternative with a minimal learning curve. My 5-year old uses it to run her legacy educational games under Win98 in Vista.

http://www.microsoft.com/downloads/details.aspx?FamilyId=04D26402-3199-48A3-AFA2-2DC0B40A73B6&displaylang=en

[Trader.Jon]

Mike,

Any less likely to have 'out of memory' issues with NT running on Server 2008?

I was considering testing the combo out and would appreciate your thoughts.

Thanks,
Jon

[WhoKnows]

Mike, looking at your setup you spend quite some attention to the hard disk side of your machine.
Your MB has a Raid Controller, but as I read it all disks are connected to a separate Raid controller, 4 SSD's in Raid 0 and 2 750GB HD's in Raid 1.

- Did you choose this setup for data security, speed in general, or does it help to speed up the optimizations? I would expect that during optimization the hard disk would not be used, keeping it all in memory, but I might be wrong. Could you elaborate on this?

- Can the diferent VMware virtual maachines run from the same harddisk, of is it faster to use more harddisks and allocate each virtual machine to a different harddisk?
I tried to find info on this and asked two shops but ....only got blank stares.

[ctrlbrk]

Mike,

Any less likely to have 'out of memory' issues with NT running on Server 2008?

I was considering testing the combo out and would appreciate your thoughts.

Thanks,
Jon

Jon,

No, it is the same no matter the platform in my experience (x86, x64, 2GB or 12GB, XP, Vista or 2008).

However, there have been some recent developments (in the last 24 hours) on the out of memory issue. I believe Dierk and his team may have made a breakthrough. It will be included in NT7 and I will let NT give the details when they are ready, but it works!

Mike

[ctrlbrk]

Mike, looking at your setup you spend quite some attention to the hard disk side of your machine.

As my friend Teal'c from Stargate would say, "Indeed."

LOL! No the main reason I am crazy about storage has to do with my last day job, I was the VP at a data storage company. We sold subsystems capable of doing 2,000MB/sec, active/active, no single point of failure, etc. So all the disk bottlenecks got to me and now all my personal systems have killer disk setups.

Your MB has a Raid Controller, but as I read it all disks are connected to a separate Raid controller, 4 SSD's in Raid 0 and 2 750GB HD's in Raid 1.

Correct, on-board is not fast enough for the SSD's in a RAID 0.

- Did you choose this setup for data security, speed in general, or does it help to speed up the optimizations? I would expect that during optimization the hard disk would not be used, keeping it all in memory, but I might be wrong. Could you elaborate on this?

Speed in general. It really has zero effect on Ninja, but this system is my primary desktop as well. Also, I am impatient and don't like to wait. Typically on any modern system, the hard drive is the slowest factor. It still is on my system, even pushing 700MB/sec. NT does keep the optimizations in memory. But typical day-to-day activity sees a nice performance boost from the SSD RAID 0. The RAID 1 is my backup disk, I have some real-time sync software that pushes critical files (ie: Ninja files) to the second RAID mirror for security and backup. Then I Acronis my entire system on a scheduled basis to the RAID 1 for security/backup/disaster recovery. All of this is overkill for most, but my background makes me do this.

- Can the diferent VMware virtual maachines run from the same harddisk, of is it faster to use more harddisks and allocate each virtual machine to a different harddisk?

Yes, most VMware or VirtualBox setups use logical files like a flat VMDK file that resides on-top of your current file system. In other words, c:\vmware1.vmdk. It's just a file. It is possible to give direct access to the LUNs and bypass the native OS, allowing the VM's OS to directly access the LUN, but that is overkill for a workstation and generally only used in a VMware ESX cluster for VMotion failover.

Long story short, you just tell the VM (or multiple VM's) what size you want allocated to them, and they create the file on your existing disk.

I tried to find info on this and asked two shops but ....only got blank stares.

Yes, not surprising Very few people really fully understand storage subsystems completely, but with the advent of low-end consumer RAID cards from Adaptec, 3Ware and Areca (ie sub $500) more and more "power users" are using and discovering them.

Mike

[WhoKnows]

Thanks, very useful info ... I guess I'll can start ordering some stuff. Too bad Newegg does not deliver in Australia.

BTW Can you believe that in the shops they don't understand you only want two cheap video cards with your I7 920... They kept explaining/complaining to me I could not crossfire something and would not be able to play games.

Just as if trading is not enough fun and excitement.

Peter
and too bad SG stopped ... Teal'c no more )

[Garyasdf]

Ctrlbrk, thanks for starting this. I'm a little surprised that this hasn't been undertaken by the engineers/coders at NT itself. Quick newbie question: considering that NT seems so graphics intensive (at least to me, granted no 3D, but still), why is it that a graphics card isn't more important? I can guess that the i7 is so new that maybe it doesn't have to pipe out any of the graphics tasks, but wouldn't it help on a lower end system?

-Gary

[ctrlbrk]

Hi Gary,

Hmm. In my opinion, NT is not very graphics intensive, these are pretty simple plots we're dealing with and the amount of redraws or refreshes ("fps") is minimal, single digit for sure.

In fact, since most high-end systems have a high-end video card, it would be nice if NT would be written to take advantage of CUDA, the ability to use all the left-over GPU cycles for something productive. GPU's can be massively more efficient at number crunching than the typical CPU, but I suspect we will never see that happen, I don't think NT is known for being cutting edge.

NT has mentioned they do their own performance tests, which they do not share.

Mike

[zoltran]

I'd have to agree with Mike ... NT isn't a graphics intensive app.. Nothing like games are.
Most standard video cards will do well.
Also .. read back a bit .. you'll see that NT dev are doing benchmarks, at a more rigrous level than we are here.

[ClayB1]

NT rewritten to take advantage of CUDA - now that's an idea

I use Acronis to image my system on a regular basis also, but the images are stored on a removable hard drive and kept in a fire safe. I'm very paranoid about my data - probably comes from working in the IT industry for so many years.

[darckeen]

One question about the benching methodolgy, how many hours of 1 min data are you using, I hadn't seen any stated standard. Just thinking I could test one day of usdjpy and be comparing it to someone who ran it on something with a much shorter trading session.

Cool thread and those are some really nice i7 systems, didn't know they overclocked so well heh. ctrlbrk what speed memory are you running, overclocked 1333 or underclocked 1666?

[ctrlbrk]

darckeen,

I am using 1600 underclocked slightly, my exact specs are in previous posts.

The benchmark script does not use any minute data, it just pushes some numbers thru NT so the input set is identical from one user to the next.

Mike

[Frankenstein]

Mike, your 'benchmark' may give a first good idea of what is possible
with Core i7 over older peers, but it should be kept in mind that this
'benchmark' is just calculating a few math-functions (tan,sin,cos,log...)
which results are never used (not very practically) and the parallism
of the loops is perfectly suited for Core i7 HT and its cache hierarchy +
the direct memory access (especially when you allow the threads to jump
through the cores in the taskmanager; .NET in Seven may bring additional
performance over older versions) whereas regarding "real-world-calculations" -
meaning operating on (cached) memory with more (cache-local) memory
operations, more integer operations and a less degree of just brute force
math calculations and parallelism, there is no real difference between
Core i7 and Core 2 at the same clockspeed (i have practially experience
with 4,0GHz and (water-cooled) 5,0GHz here... till NinjaTrader can enhance
all of the cores of course.

For most users, Core 2 is still the best $/performance alternative, especially
as there is nearly no performance difference at all.

[darckeen]

I am using 1600 underclocked slightly, my exact specs are in previous posts.

Thats darn pretty good considering you got 12 gigs. I actually thought you might have had to go a little faster then 1666 to get 12G as that speed. Very nice.

For most users, Core 2 is still the best $/performance alternative, especially
as there is nearly no performance difference at all.

I've been investigating a new box for NT7, after all the end of the quarter isn't too far away ;P I was thinking of going with a AMD Phenom II 940 but now i'm not sure...