Unable to submit / retrieve new work

Hi,

I've finished my first two work units but I'm unable to submit them at the moment, and have been since completion a number of days ago.
The output is showing the following:

[Comm thread Nov 9 01:14] Updating computer information on the server
[Comm thread Nov 9 01:14] PrimeNet error 7: Invalid parameter
[Comm thread Nov 9 01:14] parameter s: Invalid int value/precision '25'
[Comm thread Nov 9 01:14] Visit http://mersenneforum.org for help.
[Comm thread Nov 9 01:14] Will try contacting server again in 15 minutes.

I'm running XP pro 32bit. Version 26.6 build 3

The other question I have is that when I started the work initially, only enough for 2 of my logical cores. I'm running a dual HT Xeon system. When I run the stress test, all 4 workers are running, but not when doing work. Is this normal? If not, can I force the activity of all cores?

Many thanks in advance.

[Prime95]

Turn on debug info (in the Test/Primenet dialog box, go to the Connection... dialog box)

Try contacting server again.

Send the prime.log file to me (or post the relevant lines here)

[petrw1]

Originally Posted by Unregistered

The other question I have is that when I started the work initially, only enough for 2 of my logical cores. I'm running a dual HT Xeon system. When I run the stress test, all 4 workers are running, but not when doing work. Is this normal? If not, can I force the activity of all cores?
Many thanks in advance.

The topic of HT threads has been analyzed and discussed quite a bit on this forum. The bottom line is that (in ALMOST all cases) you are better off to NOT use HT for prime95. Since Prime95 will virtually run each core at 100% the HT cores rarely help and usually hurt a little.

[Dubslow]

Isn't a Xeon a quad core?

If you're talking about a dual core hyperthreaded processor, then under Test/Worker windows, you can tell it to run two threads per worker. That will make use of all four logical cores while getting maximum throughput on two workers. (Those two workers will be running about twice as fast as if you have four workers for each logical core.)

Even if it's a 4 core/8 thread proc, the above still applies.

petrw1, in my experience, running 1 worker per physical core without the helper thread is not as effective, because it's much easier for some useless process to interrupt the other logical core and use resources. In general though you are right, it is better to run one worker per physical core as long as you have the helper thread.


Also, P95, in the interim, should he manually submit his results file?

[Hop]

Originally Posted by Prime95

Turn on debug info (in the Test/Primenet dialog box, go to the Connection... dialog box)

Try contacting server again.

Send the prime.log file to me (or post the relevant lines here)

Thought I should probably register. Hi All.

Recent debug output is:

 [Wed Nov 09 11:18:06 2011 - ver 26.6]
Updating computer information on the server
URL: http://v5.mersenne.org/v5server/?v=0.95&px=GIMPS&t=uc&g=0fc820426e7bf7bf206908ba64565154&hg=0a5278a69b98470c4e82ac5dc50ab305&wg=6e9c82ebea125a192a0f30d44af390ca&a=Windows,Prime95,v26.6,build+3&c=Intel(R)+Xeon(TM)+CPU+3.20GHz&f=Prefetch,SSE,SSE2&L1=16&L2=2048&np=2&hp=2&m=2047&s=25&h=24&r=1055&ss=41&sh=0791A9304AF2B16A7D36181800563560
== Info: About to connect() to v5.mersenne.org port 80
== Info:   Trying 71.6.220.163... 
== Info: connected
== Info: Connected to v5.mersenne.org (71.6.220.163) port 80
=> Send header: GET /v5server/?v=0.95&px=GIMPS&t=uc&g=0fc820426e7bf7bf206908ba64565154&hg=0a5278a69b98470c4e82ac5dc50ab305&wg=6e9c82ebea125a192a0f30d44af390ca&a=Windows,Prime95,v26.6,build+3&c=Intel(R)+Xeon(TM)+CPU+3.20GHz&f=Prefetch,SSE,SSE2&L1=16&L2=2048&np=2&hp=2&m=2047&s=25&h=24&r=1055&ss=41&sh=0791A9304AF2B16A7D36181800563560 HTTP/1.1

Host: v5.mersenne.org

Accept: */*



<= Recv header: HTTP/1.1 200 OK

<= Recv header: Server: Microsoft-IIS/5.0

<= Recv header: Date: Wed, 09 Nov 2011 10:18:05 GMT

<= Recv header: MicrosoftOfficeWebServer: 5.0_Pub

<= Recv header: X-Powered-By: ASP.NET

<= Recv header: Connection: close

<= Recv header: X-Powered-By: PHP/5.2.5

<= Recv header: Content-type: text/html

<= Recv data: pnErrorResult=7
pnErrorDetail=parameter s: Invalid int value/precision '25'
==END==

== Info: Closing connection #0
RESPONSE:
pnErrorResult=7
pnErrorDetail=parameter s: Invalid int value/precision '25'
==END==

PrimeNet error 7: Invalid parameter
parameter s: Invalid int value/precision '25'

Hope that's of some use.

The Xeon's an older variety. I've got two physical cores on the mobo (with 2 heatsinks etc) and each one is hyperthread capable. P95 isn't quite picking up the speed correctly though. It's identified it as a 25Mhz core. My first expected completion date was mid next year. I also had to force that it was 2 physical, 2 HT in the config file as well.

I guess the 2 threads per worker would still fix this. Can't see an explicit option for that, so what line do I add to which file to get that going?

[Dubslow]

Are you saying you have two separate Xeon chips, each of which with one core and two threads?
In the Test/Worker Windows dialog box, the option you want is
"CPUs to use (multithreading):"
It's at the bottom of the Test/Worker Windows dialog box.

If you still can't find it, you'll want
ThreadsPerTest=2
in local.txt, same place as you put the NumCPUs=2 (or NumPhysicalCores=2 ; either should have the same effect).

[Dubslow]

Interesting. In my quest to understand computers better, I compared that to the PrimeNet API document, more specifically the "update computer" section, 5.1: http://v5.mersenne.org/v5design/v5webAPI_0.97.html#5.0

s - speed of CPU in Mhz; assumes all CPUs are same speed (integer)

pnErrorDetail=parameter s: Invalid int value/precision '25'

Apparently a speed of 25 MHz is not a valid value for the s parameter, which is to be expected. I couldn't tell you why it's failing to determine speed correctly, you'll have to wait for George to get back. My guess would be that your CPU's are old enough that P95 is misidentifying it as something else, then sees the lower performance and guesses s=25. Either way, you ought to be able to override this in local.txt with a
CPUSpeed=3200
line.
A quote from undoc.txt:

The program automatically computes the number of CPUs, hyperthreading, and speed.
This information is used to calculate how much work to get.
If the program did not correctly figure out your CPU information,
you can override the info in local.txt:
NumCPUs=n
CpuNumHyperthreads=1 or 2
CpuSpeed=s
Where n is the number of physical CPUs or cores, not logical CPUs created by
hyperthreading. Choose 1 for non-hyperthreaded and 2 for hyperthreaded. Finally,
s is the speed in MHz.

As an alternative to the above, one can set NumPhysicalCores=n in local.txt.
This is useful on machines that are somtimes booted with hyperthreading enabled
and sometimes without. Normally, the program can detect this situation, but one
notable problem case is a dual-CPU hyperthreaded machine, For example, take a
dual-CPU quad-core hyperthreaded machine. When booted with hyperthreading enabled
this is properly detected as an 8-core hyperthreaded machine. When booted
with hyperthreading disabled, this is improperly detected as a 4-core hyperthreaded
machine. If you set NumPhysicalCores=8, then the program will set the
hyperthreading state properly no matter how the machine is booted.

It seems both of your problems are found in the same two paragraphs. Hope this helps. Did I get anything wrong George? (If not, awesome!!!!)

[Prime95]

Originally Posted by Dubslow

Did I get anything wrong George? (If not, awesome!!!!)

Right on target. Well done!

[Hop]

Originally Posted by Dubslow

Apparently a speed of 25 MHz is not a valid value for the s parameter, which is to be expected. I couldn't tell you why it's failing to determine speed correctly, you'll have to wait for George to get back. My guess would be that your CPU's are old enough that P95 is misidentifying it as something else, then sees the lower performance and guesses s=25. Either way, you ought to be able to override this in local.txt with a
CPUSpeed=3200

I'm going to guess that 25 is the minimum allowed value.

Total pebkac situation. Why? Ready for this? I'd misread as Ghz and in local.txt put s=3.2
No wonder it's failing as an int...

All suggestions heeded. Let's try submitting....

Yep, that seems to have done the trick!

Big thanks. Back to factoring. The heaters have all packed up in my computer room so this actually makes a reasonable difference :)

[Dubslow]

What sort of factoring? If it's TF at the current LL wavefront, I'd switch to P-1 or TF-LMH; with GPU's, roughly ten of us can do all the work that CPU's once did at the LL wavefront. (Can post details on request.)

[Dubslow]

Originally Posted by Dubslow

Isn't a Xeon a quad core?

If you're talking about a dual core hyperthreaded processor, then under Test/Worker windows, you can tell it to run two threads per worker. That will make use of all four logical cores while getting maximum throughput on two workers. (Those two workers will be running about twice as fast as if you have four workers for each logical core.)

Even if it's a 4 core/8 thread proc, the above still applies.

petrw1, in my experience, running 1 worker per physical core without the helper thread is not as effective, because it's much easier for some useless process to interrupt the other logical core and use resources. In general though you are right, it is better to run one worker per physical core as long as you have the helper thread.


Also, P95, in the interim, should he manually submit his results file?

This is the most recent post I can think of about this, so I'll redirect this thread:

From http://www.anandtech.com/show/4830/i...ure-exposed/2:

Some structures within the chip are now better optimized for single threaded execution. Hyper Threading requires a bunch of partitioning of internal structures (e.g. buffers/queues) to allow instructions from multiple threads to use those structures simultaneously. In Sandy Bridge, many of those structures are statically partitioned. If you have a buffer that can hold 20 entries, each thread gets up to 10 entries in the buffer. In the event of a single threaded workload, half of the buffer goes unused. Ivy Bridge reworks a number of these data structures to dynamically allocate resources to threads. Now if there's only a single thread active, these structures will dedicate all resources to servicing that thread.

That explains what I was saying about running a single thread on an HT core. It's also nice to see Ivy Bridge will fix this!