Strange

eOn code for long time scale dynamics

Moderator: moderators

Post Reply
jon
Posts: 16
Joined: Sun Nov 10, 2013 7:38 pm

Strange

Post by jon »

On occasion one might see in the folps trace on the Home page, what i call a 'wct building(s)', a large jump in proformance, the last on this problem was around noon(L) 2 days ago. One of these jumps even looked like the wils building several large problems ago, the biggest wct ever seen was 70+P, vice the problem which seems to usually run around 2.5T to 3.5T. Oddly, Yesterday(L) at some time begin what might be called the 'mountain range' with a peak reaching 200P around 8:30PM.
On occasion one might see a spike in # spare wu, which seem to coincide with the issue of a small number of (new?) jobs, the latest spike was 3200 around at 1225(L) yesterday, the server usually has about 1900 stored. The largest spikes ever seen were during the Eon to Supercomputer compairson tests a while back. Oddly during the end of the last large problem (27 million wu) and also this latest problem, these issue of (new?) jobs, which have completion times of only 'several real seconds', started happening, the 1225(L) being the latest also.
Are these things?
jon
Posts: 16
Joined: Sun Nov 10, 2013 7:38 pm

Re: Strange

Post by jon »

OMG, 400+P plateau @ 1820(L).
jon
Posts: 16
Joined: Sun Nov 10, 2013 7:38 pm

Re: Strange

Post by jon »

New height; "1.5?" at 0810(L) CST 29 Nov 2013 duration (typical, usual) of about 1 hour, Previous only "400+P". "?", to big for a letter?
wu spike of about 3500 at 2325(L) CST 28 Nov 2013 of short durtation wus.
jon
Posts: 16
Joined: Sun Nov 10, 2013 7:38 pm

Re: Strange

Post by jon »

A couple of somewhat normal sized 1 hour P range FLOPal excurions, oddly there is also a 'gap' (no red line) in all traces (FOLPS (Hourly Average), # of Compters (Hourly Average), # spare wu), from 0355(L) to 0430(L), 04 Dec 2013(L)(CST). This 'gap' was followed in a few minutes by the # spare wu trace dropping to zero and then springing to 4700, then sabilizing slightly lower than the normal 1900 wus. (No issue of small wus seems to have occured concurrently).
The number of cores seems to be at near record, is there a optimal size of wu that runs on the dynamic collection of machines that is the BOINC-A-SPHERE that will yield the largest EON throughput.
graeme
Site Admin
Posts: 2253
Joined: Tue Apr 26, 2005 4:25 am
Contact:

Re: Strange

Post by graeme »

I think these are caused by lags on the server due to loads from the backup system. When the boinc system catches up, it records many work units as processed in a short period of time, giving an unreasonably large flops measure.
Post Reply