Server Upgrade in progress

Message boards : News : Server Upgrade in progress

To post messages, you must log in.

1 · 2 · 3 · Next

AuthorMessage
Timo Strunk
Volunteer moderator
Project administrator

Send message
Joined: 6 Sep 07
Posts: 560
Credit: 10,729,563
RAC: 0
Message 4467 - Posted: 4 Nov 2011, 10:06:42 UTC
Last modified: 5 Nov 2011, 2:04:16 UTC

We are upgrading the server at the moment. Please excuse some hiccups during the upgrade.

Edit (2:44 am) : After a long fight with the libraries and the client we are online again. Please report anything unusual, any errors, etc. in this thread. Work is now generated.
The upgrade went ok in the end. However: The current BOINC server code has a bug in the fixed credit system, which was already fixed in the trunk. For the moment we are therefore switching to the new credit system and we will further evaluate it for the next week. If it doesn't give good results, we will switch back to our fixed credits by backporting the fixes from trunk to server_stable.
We did not upgrade everything yet - there is still a hardware upgrade to be done (upgrading the storage to 8TB).

Und auf Deutsch:
Nach einem langen Kampf sind wir wieder online. Es gab ein paar Probleme mit dem Code zur Creditvergabe. Da jener fuer fixed credits im momentanen BOINC server code etwas verbuggt ist (im aktuellen trunk code ist er aber bereits gefixt), probieren wir fuer diese Woche das neue Credit System aus. Sofern das nicht die gleichen Resultate bringt, wie das fixed Credit System, werden wir die Fixes backporten und wieder zum fixed Credit System wechseln.
Alles konnten wir diesmal nicht upgraden: Ein Hardware Upgrade steht noch aus. (Wir upgraden die Speicherkapazitaet auf 8 TB).

Danke nochmal für eure Geduld!
ID: 4467 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Marie

Send message
Joined: 22 Oct 11
Posts: 4
Credit: 2,031
RAC: 0
Message 4469 - Posted: 4 Nov 2011, 17:37:02 UTC - in response to Message 4467.  

I'm not able to connect and send the done work. So I think it's normal. I'll be waiting.
ID: 4469 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Corialanus

Send message
Joined: 27 Aug 11
Posts: 13
Credit: 160,289
RAC: 0
Message 4473 - Posted: 5 Nov 2011, 11:42:40 UTC - in response to Message 4467.  

Not a very good change over.

I received over 20 work units.

They all said they were 347 Gflops and would take 49 seconds to run. I assumed there was something wrong with them.

on my computer 347 gflops would be two mins.

Why no estimated Gflops in the tasks anymore ?

I tried one task and Time remaining goes UP rather than being decremented.

Why are all new units showing 347 Gflops.

Have done one unit and the credits accrued seem to have halved !!

For several hours could not get on to my account to even report issues.

Not very impressed with this change really.
ID: 4473 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Jochen

Send message
Joined: 27 Sep 10
Posts: 49
Credit: 37,347,110
RAC: 0
Message 4475 - Posted: 5 Nov 2011, 14:16:28 UTC

The estimated running time is correct, once you have finished the first WU.

Sad but true, the credits have dropped by 50 percent, even more in some cases.

As well there seems to be a limit of 30 WUs per computer. Even less than before?!? I thought you were considering to increase the number of WUs dispatched with the new client?
ID: 4475 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Timo Strunk
Volunteer moderator
Project administrator

Send message
Joined: 6 Sep 07
Posts: 560
Credit: 10,729,563
RAC: 0
Message 4481 - Posted: 5 Nov 2011, 18:59:14 UTC - in response to Message 4475.  

Hi everybody,

The 30 workunit limit is definitely not intentional - once we find out what it was, we will revert it at least to the old system. Also the credit drop is obviously not intentional. Currently to gauge the new boinc credit system all workunits should have the same runtime. The flop count estimate in the new system is only a relative estimate - as long as a job running half the time has half flops the credits should be fine.

And if I understood the system correctly, credits should increase quite a lot in the next two days.

Best,
Timo
ID: 4481 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile dskagcommunity
Avatar

Send message
Joined: 26 Oct 11
Posts: 117
Credit: 368,495,687
RAC: 0
Message 4483 - Posted: 5 Nov 2011, 19:00:59 UTC - in response to Message 4475.  

Here all is running fine. Credits are less then before here too :/

DSKAG Austria Research Team: http://www.research.dskag.at



Crunching for my dead Dog who had "good" Braincancer..
ID: 4483 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Jochen

Send message
Joined: 27 Sep 10
Posts: 49
Credit: 37,347,110
RAC: 0
Message 4484 - Posted: 5 Nov 2011, 20:33:57 UTC - in response to Message 4483.  

Yes the runtime is about the same for all new workunits. But the runtime doubled in comparison to the old gvpj workunits.

energystats wu: 700 s, 23 credits (average)
gvpj wu: 1500 s, 36 credits (average)
barrelstab. wu: 3100 s, 31 or 44 credits (fixed)


But I now do as well have the problem with a 'remaining time' of 34 hours (slowly decreasing, currently at 26 hours). It's even worse on my Q9650. I don't know what it started with, but it's now showing a remaining time of 72 hours.

At least the new client and workunits seem to be running fine on Win XP x86 and Win 7 x64. No problems so far.
ID: 4484 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Corialanus

Send message
Joined: 27 Aug 11
Posts: 13
Credit: 160,289
RAC: 0
Message 4494 - Posted: 7 Nov 2011, 13:08:59 UTC - in response to Message 4484.  

Still no change on the units given, only getting less than half the previous on completed tasks.

Any idea when this will rectified please ?
ID: 4494 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Timo Strunk
Volunteer moderator
Project administrator

Send message
Joined: 6 Sep 07
Posts: 560
Credit: 10,729,563
RAC: 0
Message 4497 - Posted: 7 Nov 2011, 14:30:47 UTC - in response to Message 4494.  

Hi back,

At the moment, the only way to get higher credits again is reverting to the old credits system. I will ask at the BOINC mailing list regarding this.

The new system was developed to normalize between multiple projects, so that every project grants the same. So my question would be: Does POEM@HOME grant less credit than for example SETI or Einstein in the same time?

Best and thanks,
Timo
ID: 4497 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Timo Strunk
Volunteer moderator
Project administrator

Send message
Joined: 6 Sep 07
Posts: 560
Credit: 10,729,563
RAC: 0
Message 4498 - Posted: 7 Nov 2011, 14:39:17 UTC - in response to Message 4494.  

Hi everybody,

I wrote to the BOINC mailing list to investigate the credit behaviour. It seems there were some more projects, where the credits dropped after the change to the new system. Hopefully the BOINC framework developers will have an answer.

Best,
Timo
ID: 4498 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Corialanus

Send message
Joined: 27 Aug 11
Posts: 13
Credit: 160,289
RAC: 0
Message 4508 - Posted: 7 Nov 2011, 17:07:50 UTC

Have checked other projects and done a view of Cpu seconds per unit of credit.

LHC@home needs 173 seconds CPU per unit of credit
SImap 138
Docking 100
Rosetta 144
Poem (new) 139

ID: 4508 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
FrankHagen

Send message
Joined: 10 Mar 08
Posts: 87
Credit: 67,454,067
RAC: 0
Message 4509 - Posted: 7 Nov 2011, 17:56:15 UTC - in response to Message 4508.  

Have checked other projects and done a view of Cpu seconds per unit of credit.

LHC@home needs 173 seconds CPU per unit of credit
SImap 138
Docking 100
Rosetta 144
Poem (new) 139


OMFG!

this is exactly what's needed now, but i should have known this would happen. :(


you can get 100 seconds on my oldest cruncher:

GenuineIntel x86 Family 5 Model 2 Stepping 12 119MHz [x86 Family 5 Model 2 Stepping 12]

boinc benchmark: Measured integer speed 141.73 million ops/sec


or one of my current ones where you see that:

boinc benchmark: Measured integer speed 14512.67 million ops/sec


if i get 1 credit for 100 seconds for the ancient PENTIUM 1, i should stop buying, should i?
ID: 4509 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Vespasian

Send message
Joined: 21 Aug 11
Posts: 1
Credit: 1,010
RAC: 0
Message 4510 - Posted: 7 Nov 2011, 18:18:34 UTC

Not quite sure what you are inferring.

My objective in supplying the figures was to illustrate the differing credits between projects.

Clearly processors are different therefore the only true unit of measure across different processors would surely be gigflops not CPU time. But as Gflops NEVER appear in the account list what other measure would you suggest !!!
ID: 4510 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
FrankHagen

Send message
Joined: 10 Mar 08
Posts: 87
Credit: 67,454,067
RAC: 0
Message 4511 - Posted: 7 Nov 2011, 18:30:57 UTC - in response to Message 4510.  

Not quite sure what you are inferring.

My objective in supplying the figures was to illustrate the differing credits between projects.

Clearly processors are different therefore the only true unit of measure across different processors would surely be gigflops not CPU time. But as Gflops NEVER appear in the account list what other measure would you suggest !!!


first - i was talking about mips - not flops. different story you know?

second - of course the credit-thing before "creditnew" was based on benchmarks (which were silly enough btw.).

and really not the last about this - look here:

4 Nov 2011 | 15:34:56 UTC Completed and validated 871.66 864.65 443.76

that's a credit in 2 seconds for pure cpu-work - creditnew on another project btw..

want a list of projects that tried it and produced havoc in the whole credit system?
ID: 4511 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Corialanus

Send message
Joined: 27 Aug 11
Posts: 13
Credit: 160,289
RAC: 0
Message 4512 - Posted: 7 Nov 2011, 18:43:22 UTC

Thanks for that I am well aware of the difference between MIPs and FLOPS .

Yes please provide the projects list.

What does btw mean ???
ID: 4512 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
FrankHagen

Send message
Joined: 10 Mar 08
Posts: 87
Credit: 67,454,067
RAC: 0
Message 4513 - Posted: 7 Nov 2011, 18:54:15 UTC - in response to Message 4512.  
Last modified: 7 Nov 2011, 19:01:42 UTC

Thanks for that I am well aware of the difference between MIPs and FLOPS .

Yes please provide the projects list.

What does btw mean ???


by the way..

AQUA - D.A. told them it would need a few days to swing in - it did - for a short time and then granted WU's with millions of credits.

DistrRTgen - same story - except this is mainly a GPU-project.

DNA - same story - travis (if you know whom i mean) had a bunch of days of work until he decided to modify.

currently numberfields is going crazy.

so anyone really has trust in it working now and here just by chance?
ID: 4513 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Timo Strunk
Volunteer moderator
Project administrator

Send message
Joined: 6 Sep 07
Posts: 560
Credit: 10,729,563
RAC: 0
Message 4514 - Posted: 7 Nov 2011, 22:11:23 UTC

Hi everybody,

I hope these projects all had the same problem:

They reacted on the complaints. If we would adjust the FLOPS now, the new credit values would be based on that and we would be immediately granting either 0 or a million credits.

From the benchmark values shown above (thanks Corialanus) it seems that the New Credit is achtually working and did swing in.

Best,
timo
ID: 4514 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Jochen

Send message
Joined: 27 Sep 10
Posts: 49
Credit: 37,347,110
RAC: 0
Message 4518 - Posted: 8 Nov 2011, 2:51:48 UTC

And it starts all over again with every new WU-type? I'm down to 26 credits per WU with Poem++ 0.08... And the estimated running time is back to a couple of days per WU again.
ID: 4518 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
BarryAZ

Send message
Joined: 12 Nov 08
Posts: 94
Credit: 171,936,261
RAC: 0
Message 4519 - Posted: 8 Nov 2011, 3:21:36 UTC
Last modified: 8 Nov 2011, 3:45:16 UTC

Timo, regarding the credit hit -- that tracks with the experience of others in NewCredit land. Further, some folks have reported that when a project optimizes their application code, the credits reduce further to match. It seems to be something of an ideological choice back at BOINC central. A number of projects, having observed it, have reverted to a manual fixed credit override.

I don't know that I should speculate as to the reasons for this choice being pushed out in the BOINC server code. However, you may find that you end up having to make a choice in responding to your user community or to the BOINC developers.

You may recall several months ago that I queried here regarding whether or not POEM was NewCredit -- that was during the back and forth which caused such a mess for Aqua (no longer running) and a fair amount of chatter in other projects (not just SETI which of course would always follow BOINC developer dictates).
ID: 4519 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
FrankHagen

Send message
Joined: 10 Mar 08
Posts: 87
Credit: 67,454,067
RAC: 0
Message 4520 - Posted: 8 Nov 2011, 6:44:25 UTC - in response to Message 4518.  

And it starts all over again with every new WU-type? I'm down to 26 credits per WU with Poem++ 0.08... And the estimated running time is back to a couple of days per WU again.


that's by design..
ID: 4520 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
1 · 2 · 3 · Next

Message boards : News : Server Upgrade in progress


Copyright © 2017 KIT-INT