Can someone Explain to me a SMP issue ?

[DPRGI - Marzio]

Hello guys,
I'm Marzio from DPRGI Team. I have a problem with the multi installation of SOB.
The platform is a P4 2.8@3.6 ghz in HT .
If i run a single client (on service) the standard cEMs/sec is 890.000 and the pc show me only the 50% of cpu use.
If I run 2 services the cEMs/sec are 490.000 each client with the 100% of cpu use.
I don't remembere that the client have smp instructions , what's happen ?

p.s.= i used sobservicecfg for make the client service configuration


thank you for help

Marzio

[Mystwalker]

As you are using hyperthreading, one program can still take next to 100% of your computing power although the task manager only shows 50%.
Both virtual CPUs share the same computing units. If one application is optimized to take almost all advantages (e.g. computing units) of a computer architecture (e.g. P4), then there is next to no gain in using a second instance.
In your example, it's even worse, as the second instance also needs resources like memory bandwidth.

So basically, your system is used to full capacity. It's just that Windows treats your system as a real SMP one...

[DPRGI - Marzio]

Many tnx Mystwalker.
I don't understand why exist programs for run more than one instance and why , in the readmeNT.txt there are more instruction for setup more than one instance in P4 H.T sistems.
the option sobsvc -o2 isn't necessary, but the readme speak about systems with Hyperthreading....

i don't understand what I must do for optimize my configuration...

[Mystwalker]

Maybe a second instance is performing better on other hyperthreaded machines...
As MathGuy can't be sure this is definitely not the case, it's nice to have the option, isn't it?

[MathGuy]

That's probably about right for a one-processor (two virtual CPU) machine with hyperthreading: two clients gives a slight overall increase (980 vs. 890) over one client.

The main place that hyperthreading helps SB is with a two-processor (four virtual CPU) machine. Three clients can be used there with (often) a significant advantage over two. This ratio appears to be close to optimum, but of course we can't run 1.5 clients on a single-processor machine.

I wouldn't say that there is anything you need to do to optimize your machine further...a lot of people would really like to have a machine doing ~900 ceM/sec! My guess is that the system is a little less responsive in the two-client configuration. It's simply your decision as to whether you like the few extra cEMs or the foreground response time better...

[DPRGI - Marzio]

don't worry MathGuy , my systems are dedicate only for D.C. , a reactive Pc is now in front of me and is dedicate for internet surfing and seti@home project.

thank you very much for the explanation