SOB: Those of you that run SOB on SMP systems ...

[MereMortal]

I posted an observation about running the SOB client(s) on SMP systems over at the official forum, but have yet to receive any useful info --- so I thought I'd try here too. My problem (or what I see as a problem, anyway) is that I see a performance hit when running multiple clients on dual processor systems. This happens with both the linux client and the windows service client.

The symptoms are as follows. I start one client and record its rate. If I then start another client in a separate directory (or add another processor to the windows service client), the rates of both clients are lower (sometimes appreciably lower) than running just one client. Has anyone here seen this phenomenon?

Practically all of my systems are SMP systems, so this is really frustrating! :disgust:

 

[deerslayer]

I haven't noticed it....but I could check into it once I get moved into my dorm room Thursday.

I am running the newest version of the client with my 1900+ dually

 

[Baldy18]

My 1800+ dually I believe (don't quote me on the single process rate) runs a single process at 240KcEM/s and with two processes they are each about 210KcEM/s. I contribute it to a FSB bandwidth bottleneck but that is really just a guess.

I'll go test it out and see if my quick guesstimates are accurate.

 

[Baldy18]

Ok I'm back and here is what I found:

2 instances:
218KcEM/sec each

1 instance:
261KcEM/sec

 

[Allio]

2 instances on a dual pentium pro @ 200MHz = 24k cems/s per cpu

Don't ask me to find out what it is for one instance, it takes about a day for it to get up to that rate

 

[MereMortal]

Baldy18, that result is in line with what I tested this morning for a dual 1800+ (275k --> 218k). This adds about 4 minutes/block.

I could probably learn to live with that. But I have access to dual Xeons, and here the linux client goes from 7.33 minutes/block to 13.5 minutes/block. That is completely unacceptable. Perhaps I'm making too big a deal about this, but I have a good deal of power that's currently going to waste.

 

[Baldy18]

Does anyone know how FSB intensive SOB is?

One other thing I didn't look at is the CPU usage while running one process. Is it useing a portion of the second CPU?

Also what rate does a single CPU 1800+ system get?

If we can answer these questions we can get closer to figuring out the cause of this.

 

[OberonBob]

I believe that SOB is memory bus limited in all most all computers. case in point: I have a AMD 1900+ (1600MHz) at home. It ran at about 210kcem with my oldish Via 266a MB. I switch to a brandnew Nvidia MB wth dual memory channels, still running at 266 and the same memory modules, same CPU, fresh install of XP, and now I am getting right at 300Kcems.

Another instance. We have many many 2.4 Gig P4s 533 FSB at work, single 266 MHz DDR bus. they do about 250kcems. We have 10 dual 2.4 Gig Zeons (400 FBS) with the same 266 memory, and they do about 310 with three clients running(hyper threading). They run about 300 with 2 clients. So, I believe it is all about the memory bus. The new Workstation we are getting at work about going to be the same 2.4 chips with dual 333 MHz ram, and I fully expect the scores to double. I will let you all know.