Wow, some great comments!
Thanks to everyone who has replied to this.
Something to keep in mind on this: I'm not representing this data as
authoritative, nor as a "help me fix this performance problem". I'm
presenting it only as a topic of interest. It seemed to touch a note
with me.
To address some of the replies:
First off, there is only so fast setiathome will go. And 8 hours per set
is about the speed limit on anything but Sun sparc stations, running Sun
OS (for some strange reason, they're faster, but that's a whole
different conversation!). I've run it on Enterprise level HP servers,
IBM SP/2 Complexes, and PCs, and found very little variation on the
per-cpu processing time (NOTE: Since setiathome is single threaded,
runing on an SMP system is faster only because you run more than one
instance...). I'd not seen a large difference, until I fired it up on
this PIII (and a few others around here).
So, having said that. A brief comparison:
The SMP PII 266.
Running Linux (Redhat 6.2).
Disks:
integrated SCSI Fast Wide/Ultra. (adaptec).
Memory
128 MB
Average Processing speed: 8 hours per cpu (2 instances running)
PII 400 UP
Disks
IDE, whatever I had handy to throw in it.
Memory
256 MB
Operating systems:
Win2k, Win98, Linux (Mandrake 8.0), and Solaris 8.0 (busy box...)
Average processing speed: 8 hours (regardless of OS. I only see minor
variations is processing speed between the different OSs. The fastest
being Linux & Solaris).
The PIII 650 Laptop UP
Disks:
Whatever Dell threw in it at the factory (IDE)
Memory:
128 MB
Operating System:
Win98
Average processing speed (I checked this): 72-80 hours.
I only have it ruinning as the screen saver on the laptop, but it's in
that mode 16 hours a day, and all week-end. That's not a reason for the
slow performance. It's doing NOTHING else at that point. I've also
disabled the power management, so it's not shutting down, or sleeping..
I thought of that already.
Nearly every machine I've seen runs these around 8-12 hours each. But
this one. Nor some other PIII machines around here.
I reflected on this, only because I was seeing people reporting slow
performance with KDE (among other things) with PIII systems, and it
sparked a note with me. Could be a problem with the PIII's, or there is
code in the newer Linux that "some" PIII's just don't process well.
I'm only presenting this as a point of interest. I may present this to
the satiathome list, and see if I can get any others witnessing the same
performance.
(the fastest I've gotten setiathome to run:
96 sets a day. But, that was on a 32 processor cluster. :) But the
average speed of seti was still about 8 hours per set, per processor.
s wrote:
> I don't know, there are alot of depends...
> I have a PIII 733 that puts out a work unit in about 12 hours, and a PIII 667
> that puts em out in about 11 and 1/2. Thing is the 733 has 256 ram and a
> 7200 rpm hdd, while the 667 has 512 ram on a 5400 rpm hdd. So my point is
> while your comparing cpu don't forget to factor in the amount of ram, bus
> size, and hdd mb/s and other processes running.
>
> -s
>
>
> On Wednesday 23 May 2001 05:26 pm, you wrote:
>
>>I thought this was interesting.
>>I have an "older" 400 PII, and a "newer" 650 PIII (laptop).
>>Using Setiathome as a performance curve to compare by, there is
>>something amiss with PIII processors.
>>
>>My 400 PII will complete a "set" of Seti data is an average of 8 hours
>>(my dual 266 PII will complete one set every eight hours, per CPU, so it
>>pumps out 2 in 8 hours).
>>The PIII takes 50 hours! per set.
>>
>>Just a data point. The PIII is a laptop, on a different OS. But if I
>>boot my 400 PII in that OS, it will still complete them in 8 hours, vs.
>>50 on the PIII
>>
>>I find this very interesting. While the PIII shows some performance
>>abilities that the PII does not (it multi-tasks better, and smoother),
>>there are some things that the PIII seems to just suck at!
>>
>>What's just as interesting is that it's not consistant. There are
>>similar machines areound me here at work. Some scream, some crawl. All
>>with similar set ups. It has me wondering if Intel has a QA problem?
>>
>>This could be an indication of why some people are reporting performace
>>problems with Mandrake 8.0 / KDE, and others are not.
>>
>>Just a thought. :)
>>
>>Ric
>>
>
>
