highFreq or lowFreq

Dario666
Dario666
Joined: 23 Dec 12
Posts: 1
Credit: 14,222,249
RAC: 0
Topic 207795

Hi

Why are some computers only counting highFreq tasks and not being able to download lowfreq tasks? Other computers only counting lowFreq tasks and not being able to download highfreq tasks. What is going on?

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1,364
Credit: 3,562,358,667
RAC: 0

The amount of memory used

The amount of memory used varies by frequency, from a few hundred MB at the low end to two GB at the high end; the scheduler splits computers into low and high end buckets and sends low frequency tasks to the former and high frequency to the latter.

MarkJ
MarkJ
Joined: 28 Feb 08
Posts: 437
Credit: 139,002,861
RAC: 2

In the last round of Gravity

In the last round of Gravity wave work they also split the work into two types. They used CPU cache memory and available RAM to determine which machines could do what.

As mentioned by Danneely the Lo work units are sent to less capable machines. The Hi go to the more capable machines. The Hi work units can use up to 1.8GB of memory each depending on the work unit frequency (above 1420Hz).

Nick
Nick
Joined: 12 Oct 13
Posts: 27
Credit: 8,949,649
RAC: 0

I have a high capacity

I have a high capacity machine and only see low units.

 

Richie
Richie
Joined: 7 Mar 14
Posts: 656
Credit: 1,702,989,778
RAC: 0

Nick_43 wrote:I have a high

Nick_43 wrote:
I have a high capacity machine and only see low units.

"Measured floating point speed:3264.54 million ops/sec"

I'm not sure if that FP speed is considered, but maybe that is not fast enough.

archae86
archae86
Joined: 6 Dec 05
Posts: 3,157
Credit: 7,206,734,931
RAC: 943,082

Nick_43 wrote:I have a high

Nick_43 wrote:
I have a high capacity machine and only see low units.

While the intent is to distribute by capability, I think the actual distinction is made rather simplistically from something as simple as the text string reported as the CPU model, with the matching code built up from early trial experience.  I agree it does not always match actual capability.

Christian Beer
Christian Beer
Joined: 9 Feb 05
Posts: 595
Credit: 183,969,759
RAC: 348,381

The distinction between the

The distinction between the two application is indeed made by the CPU model that is reported by the BOINC Client. You can read more about why this was necessary here: https://einsteinathome.org/content/gravitational-wave-search-o1as20-100-f-and-i-faq

I monitored per task runtimes for both searches and they are in the expected 8-10h range for the majority of hosts in each search.

Nick
Nick
Joined: 12 Oct 13
Posts: 27
Credit: 8,949,649
RAC: 0

I would think my 8MB of cache

I would think my 8MB of cache would be considerable, maybe it isn't these days...

 

 

archae86
archae86
Joined: 6 Dec 05
Posts: 3,157
Credit: 7,206,734,931
RAC: 943,082

Nick_43 wrote:I would think

Nick_43 wrote:
I would think my 8MB of cache would be considerable, maybe it isn't these days...

That's a Nehalem EP which was quite a fearsome chip when new.  It is no longer new, nor even middle-aged.  I've just retired my system which was running a Xeon E5620 Westmere, which is a quick redraft of Nehalem on the next generation manufacturing process (32 nm down from 45 nm for yours).

The proof of the pudding is not in any single parameter, but in elapsed time to finish a WU of a given type.  You appear to be generating valid 1Spot1TLo work with elapsed time in the 40,000 to 50,000 second range.  That is not markedly fast.

For comparison, my Skylake laptop, which is running a single instance of THi is finishing them in about 34,000 seconds.  I consider that surprisingly fast.  Perhaps this application likes Skylake for some reason.

My Sandy Bridge desktop, my oldest model in continuing service, is being giving Thi work and finishing them in about 33,000 seconds.  Again this is single instance.

My Haswell desktop, which is more modern than the Sandy Bridge model, but an i3 variant, which may distract the selection method, is getting TLo work and finishing them in about 25,000 seconds, also on single instance (but supporting two GPUs, which keeps the two physical cores of that chip pretty well occupied).

My Devil's Canyon desktop which is my most capable machine, but a generation less modern than my Skylake laptop, ran some THi work in about 32,000 seconds, but I stopped that work type the better to support my two most capable graphics cards on that host.

Nick, 

Boinc reports your host as having 16 processors.  But Intel lists that CPU as a four physical core part with hyperthreading.  Are you running a motherboard with two CPU sockets?  Plus have hyperthreading turned on?

 

Nick
Nick
Joined: 12 Oct 13
Posts: 27
Credit: 8,949,649
RAC: 0

ARCHAE86 Boinc reports your

ARCHAE86

Boinc reports your host as having 16 processors.  But Intel lists that CPU as a four physical core part with hyperthreading.  Are you running a motherboard with two CPU sockets?  Plus have hyperthreading turned on?

Yes to all of the above.

 

AgentB
AgentB
Joined: 17 Mar 12
Posts: 915
Credit: 513,211,304
RAC: 0

The X5570 also does not

The X5570 also does not support AVX, so that may also be a factor.

 

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.