Moo! Wrapper News

<a href="http://moowrap.net/">Project Link</a>
<a href="http://moowrap.net/team_display.php?teamid=465"><font color="brown">Team Link</a>

Moderator: the silver surfer

Moo! Wrapper News

Postby the silver surfer » Tue May 17, 2011 6:39 am

ATI/nVidia requirements

Hi,

Project now enforces following ATI/nVidia requirements. Most notable change is that CC 2.0 and above are no longer rejected for CUDA. This means devices like GTX 465 are now offered the CUDA application. We'll see how things works for them.

There's also better logging (for me) and also a generic notification to new BOINC Clients when these requirements are not met.

ATI Stream requirements:


BOINC Client v6.10.17
Driver v10.4 or later
Minimum device memory 250M
ATI Runtime (not older AMD)



nVidia CUDA requirements:


BOINC Client v6.4.5
Driver v256.00 or later (recommendation is at least v257.21, which is latest for this family of drivers)
Minimum device memory 384M
CUDA Runtime v3.1 or above
Compute Capability 1.0 or higher (basically all CUDA capable cards)



I probably should find a place on the website to list these requirements. They might change as we get more information about what are really required.

-w 3 May 2011 | 8:13:55 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Tue May 17, 2011 6:40 am

Credits

Hi,

I've been working on getting the crediting to stabilize. During these changes there was a time where valid results were given a near zero credit (it was shown as zero). This affected 290 results for 54 users and 80 hosts.

I've just fixed credit for these anomalies by granting them a fixed amount. Also, the near zero credits shouldn't happen anymore since it seems to have been result of a --max_granted_result switch for the validator that I tried as a solution for the credits. I'm not sure if it's a bug in BOINC Server code or if it's really supposed to work that way..

Otherwise, the crediting seems to have somewhat stabilized but I think we are giving too much credit at the moment. I'm going to bring it down a bit.

There are still cases were huge credit is given for some results. If validator doesn't correct itself I need to change how we credit results and bypass all these scaling that goes on.

-w 5 May 2011 | 10:14:13 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Tue May 17, 2011 6:42 am

New application version deployed

Hi,

I just released v1.01 of our wrapper application. In addition to previous 32-bit Windows versions there's now versions for 64-bit Linux. Both Stream and CUDA are available for both.

Some of the major changes in this release:


Overactive hang detection code fixed and it now takes into account relative speed of the used card. This should fix many failures and even slowness seen by users especially on CUDA platform.
Limited retries (no checkpoint or premature exit) to ten (10) so that there's no chance for an endless loop.
Added support for setting core used by the client. I'll add a preference for this soon. Allows advanced users to set best performing core based on their knowledge or client benchmarks.
There's now a default Distributed.net ID used if one is not specified in preferences. For now it's simply "u_<userid>@moowrap.net" where <userid> is your User ID of your account on this project. Yes, you can still set a custom one in your preferences.
And some other minor improvements and fixes around the code.



-w
6 May 2011 | 19:58:20 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Tue May 17, 2011 6:44 am

Switched to granting static credit

Hi,

Due to numerous inconsistencies in the BOINC credit calculation, I've switched to granting static amount of credit based on stat units in a wu. Stat units in a particular wu is the last number in its name (second to last in a task name) and measures the relative amount of work in a wu.

At the moment we are giving 5cr per stat unit, which gives a little over 2kcr per current WU size. (For example, a standardish wu with 448 stat units gives 2240cr) We probably switch to 7cr eventually, which gives about 3kcr per current WU size (same 448 now gives 3136cr regardless of how long it took).

Partial result crediting works the same way it used to.

-w 12 May 2011 | 17:39:16 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Tue May 17, 2011 6:47 am

Latest ATI/nVidia requirements

ATI Stream requirements:


BOINC Client v6.10.22 or later (v6.10.17 is enforced)
Driver v10.4 or later
Minimum device memory 250M
ATI Runtime (not older AMD)
Disable CrossFireX for performance reasons



nVidia CUDA requirements:


BOINC Client v6.10.22 or later (v6.4.5 is enforced)
Driver v256.00 or later (recommendation is at least v257.21, which is latest for this family of drivers)
Minimum device memory 384M
CUDA Runtime v3.1 or above
Compute Capability 1.0 or higher (basically all CUDA capable cards)



Changes:
16. May 2011: 1) BOINC Client v6.10.22 required because it's the first version that has coproc information available to applications. (Not yet enforced.) 2) It's recommended to disable CrossFireX for ATI because it might cause a major performance degradation.

-w
15 May 2011 | 22:33:18 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Thu May 19, 2011 10:31 am

Application version 1.2 deployed

Hi,

I've deployed new application v1.2, with following major changes:


Distributed.net ID and core preferences were not used on Linux.
Default D.net ID was not used if a custom was not specified.
Changed default D.net ID to <username>@moowrap.net, where <username> is your Moo! Wrapper account name with certain invalid characters removed.
Use our domain if there's no domain (missing @) in a custom D.net ID.
Added a 15min minimum to checkpoint interval.


-w
18 May 2011 | 15:26:53 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Sun May 22, 2011 6:10 pm

Different workunit sizes added

Hi,

I just completed adding different workunit sizes and scheduler should now sent you work based on the measured speed of your host. Additionally, scheduler sents workunits that are better "match" for number of cards your host has. This should minimize idle cards at the end of wu.

I'm pretty sure there's still some tweaking to do and I will be watching how the scheduler performs tomorrow. Please, do let me know if there seems to be something odd with workunits given to your host and especially if you are now unable to get any work. Thanks!

-w
22 May 2011 | 12:39:09 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Fri May 27, 2011 8:53 am

CPU applications deployed and CUDA memory requirements changed

New CPU applications deployed for 32-bit Windows and 64-bit Linux. Windows version should also run on 64-bit Windows systems. So you should now disable CPU applications on project preferences if you don't want to use your CPU for this project.

There's a known problem with checkpoint interval (uses a default 2h interval) and core selection (selected core is shared with ATI/nVidia on the same host) for these applications. Both problems should be fixed in next application version so you should wait for that if you have a problem due to either limitation.

Additionally, CUDA requirements were changed to accept cards with only 64MB of memory. 26 May 2011 | 12:41:04 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Tue Jun 14, 2011 6:04 pm

MAC applications deployed

New MAC applications deployed with both CPU and CUDA 3.1 support. There's a separate CPU application for all three CPU variants (PowerPC, Intel 32-bit and Intel 64-bit). For CUDA 3.1 there's only Intel 32-bit application available (might work on Intel 64-bit too) and BOINC Client needs to already detect your nVidia card correctly.

This deployment is still based on application v1.2 so any known problems from Linux version are most likely still there. 14 Jun 2011 | 16:28:01 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Fri Jul 08, 2011 11:24 am

Donations are now accepted using Paypal.

You can go to http://moowrap.net/donations.php or use the link on our home page to help with our monthly operation costs. We would like to cover half of our costs through donations, which is 100 USD per month. Other half is privately sponsored by the project administrator (or covered through discounts and other such means). For details, see the donations page. Thanks for your help, every dollar counts!
8 Jul 2011 | 10:59:49 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Fri Aug 05, 2011 7:16 pm

Unexpected downtime

It seems we had our first unexpected downtime yesterday. :( We've had some offline times previously but most of them have been for a shorter period of time and on purpose due to maintenance and updates.

Access to our services were failing on Thursday from about 4:00 to 20:30 when I finally brought things back online. Those times are on my local EEST+3 timezone, which means from 1:00 to 17:30 UTC and from Wed 18:00 to 10:30 PDT. This is about 16 and half hours of lost time.

Looking through logs, this seems to have been caused by the server running out of memory and subsequently OOM-killing itself to death. I have few things I can do to prevent same problem bringing us down in the future. (Like moving the DB to a different server as the OOM-killer chose the poor DB to die on first round.)

I did notice the problems in the morning but due to unrelated complications (non-project ones) I didn't manage to get the server back online until that evening. I do apologize for this extending our downtime. :(

Everything should be back to normal now but due let me know if there are still problems around. Thanks and now let's crunch hard to make up for the lost time! ;)
5 Aug 2011 | 10:33:13 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Sun Sep 04, 2011 12:34 pm

Another unexpected outage

There was another out-of-memory event that affected our main (and only) server to bring the project down for about six hours last Friday. Services were fully down from 21:00 to 3:30 local time (EEST+3, so that's 18:00 to 0:30 UTC and 11:00 to 17:30 PDT).

Anybody who is interested reading the long technical details, please see the http://moowrap.net/forum_thread.php?id=113 forum post. Thanks for crunching!
4 Sep 2011 | 3:18:16 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Wed Oct 26, 2011 5:57 pm

BOINC Server code updated

Just finished an update of the BOINC Server code to the latest available. This brought us about three months worth of fixes and new features from the upstream developers. Most notable fixes:


web almost fully translated/-able and
top GPU model statistics (http://moowrap.net/gpu_list.php).

As always, please let us know if you find something broken. Thanks!
26 Oct 2011 | 17:01:49 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Thu Dec 01, 2011 7:15 pm

Read-only replica DB deployed

There's now a replica DB used that's used for certain heavy read-only DB operations (top statistics, server_status, lists of user workunits/tasks etc) and should take a bit load off the main DB and by doing so, help the scheduler and other critical processes do their job.

Please, do let me know if there's something that no longer seem to work. Thanks and happy crunching!
30 Nov 2011 | 14:34:02 UTC Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Thu Dec 15, 2011 2:23 pm

2% done, 98% or 9 to 137 years to go

RC5-72 project of distributed.net (where we get our work from) announced a while back that they had completed 2% of the keyspace we are checking. They also calculated that there's still more work for 9 to 137 years so we don't have to worry about running out work just yet.

For the announcement itself, please read http://blogs.distributed.net/2011/11/27/18/26/bovine/.

Thanks for crunching, keep it up! :) How about we try to push the time to get to 3% under a year? With your help, we can do it!

15 Dec 2011 | 6:46:03 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Sat Dec 24, 2011 9:01 pm

Happy Holidays!

Santa came by and demanded that I do something about recent credit levels. I obviously said no since we are already granting such high levels to begin with but he wouldn't take no for an answer. Now there's some Santa Magic in effect that makes our validator grant double credits for everybody! I'm so sorry about this and I'll fix it as soon as I have time to figure out what Santa did. :(

In other news, base credit is now 9 per stat units (the last number in wu name) and fragmented work will get a 20% bump in compensation. A fragmented work is one that has more than twice the normal amount of packets in it (the second to last number in wu name). Both of these conditions are in effect for now but I might adjust values after seeing how things progress.
24 Dec 2011 | 15:32:05 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Tue Jan 17, 2012 2:08 pm

Application v1.3 deployed

New application v1.3 has been deployed. It's first available on Linux but rest of the current platforms will follow as deployment progresses. There's going to be few new platforms as well, namely 64-bit Windows CPU and 32-bit Linux CPU/ATI Stream.

Main fixes are for a Linux crash during suspend and checkpoint/hang detection improvements. There's also support for setting a separate ATI/NV/CPU core for a host, which you can now set in project preferences at http://moowrap.net/prefs.php?subset=project.

Note that there's no urgent need to abort tasks on older applications. Work from them will still be accepted and validated normally.

For a detailed changelog, please see post http://moowrap.net/forum_thread.php?id=206.
16 Jan 2012 | 21:27:06 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Sat Jan 28, 2012 11:02 am

Planned database mainenance completed

Finished a planned database maintenance few hours ago, where our primary DB was moved to a separate host. Service levels should be returning to normal.. or actually they should be getting better now that this change is done. So no more transitioner backlog or strange scheduler failures. :)

Downtime was a bit longer than expected (started around 1:00EET/23:00UTC/14:00PST) because moving over 30G of data actually takes a while and then there was complications while making our Python based backend and PHP based web use SSL when connecting to the new MySQL server.
27 Jan 2012 | 16:10:07 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Sun Feb 12, 2012 10:22 pm

Hard drive failure on our primary database server

Our shiny new primary database server, that's been responsible for the nice performance lately, decided that things have been way too stable. So this Sunday morning at about 6:38 EET* the server killed it's primary hard drive bringing everything to a grinding halt. :(

I've switched to using our replicate DB until data center staff can replace our failed hard drive and/or server. I'm currently bringing the services back online slowly to catch things up. Note that things might be slower until the first onslaught of clients reconnecting is over.

Good news is that there shouldn't be more than few seconds of DB changes lost because our database is replicated to the secondary server. Please, do tell if you see something strange. Bad news is that there's going to be a maintenance break in the near future when I switch primary DB back to the resurrected server (maybe next weekend, if things run fine with only one DB server).

*=That's 5:38 CET or Sat 20:38 PST and for other timezones, please see http://www.timeanddate.com/worldclock/f ... 101&sort=1.
12 Feb 2012 | 18:33:28 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Postby the silver surfer » Tue May 01, 2012 9:09 pm

Domain expiration mishap

The Moo! Wrapper domain moowrap.net was inaccessible for few hours because the domain expired yesterday (30. Apr). I failed to renew it on time because of some miscommunication with my service provider representative and me been too busy with my day job to handle this correctly, on time.

Domain is now renewed for next three (3) years and should be accessible once again for everybody (or slowly getting there as DNS record updates make their way through the net). I'm sorry about this short hickup on the project availability.
1 May 2012 | 15:11:51 UTC · Kommentar
Kurt
Image
Image
Scientific publications ImageImageImage
the silver surfer
V.I.P.
V.I.P.
 
Posts: 3648
Joined: Mon Apr 21, 2008 4:38 pm
Location: In front of my screen

Next

Return to Moo! Wrapper

Who is online

Users browsing this forum: No registered users and 1 guest

cron