Lots of strange fings.

[Rockinfroggi]

Since updating to Boinc 5.8.16 some time ago nothing seems the same.

I used to get a good stock of WUs for most projects now get one or 2 at a time, uploads and downloads are forever getting stuck and often finished WUs won't report even when updating.

The only thing that gets things moving is to exit Boinc and restart it, sometimes 4 or 5 times a day.

Also, on a number of occasions I've had projects suspended and set to now more work and out of the blue they will download WUs and on checking the projects page they have been set to allow work.

I use Bam and nothing has been changed in my preferences.

Is this something that can happen over time with Boinc getting cluttered with junk? Should I run my WUs down delete boinc and reinstall and has anyone else suffered similar?


Calamity Cruncher.

[Timby]

Yeah certainly have mate, 1 machine I use at work only downloads a maximum of 2 WU's on a single core machine? I don't touch these things for weeks so its nowt I have done, so I guess its an issue with Boinc Manager?

[Rockinfroggi]

It's pretty much the same on all my systems and I scan, clean and defrag every system at least once a week.



Gary.

[UBT - Timbo]

Since updating to Boinc 5.8.16 some time ago nothing seems the same.

I used to get a good stock of WUs for most projects now get one or 2 at a time, uploads and downloads are forever getting stuck and often finished WUs won't report even when updating.

The only thing that gets things moving is to exit Boinc and restart it, sometimes 4 or 5 times a day.

Also, on a number of occasions I've had projects suspended and set to now more work and out of the blue they will download WUs and on checking the projects page they have been set to allow work.

I use Bam and nothing has been changed in my preferences.

Is this something that can happen over time with Boinc getting cluttered with junk? Should I run my WUs down delete boinc and reinstall and has anyone else suffered similar?

Calamity Cruncher.

There are a few things that may be relevant:

1) - The connection via the ISP gets snarled up.
This can affect the upload and download of WUs and hence BOINC is left twiddling it's thumbs coz it cannot get to what it needs.

You can check the message logs, but they don't really tell you about the quality of the connection - so if something fails, there isn't a good report you can get which says why (and what you can therefore do to fix it.

2) BOINC uses it's own "metrics" to decide when it needs more work.
Over time the core BOINC program has been changed a bit so that it can be a little more "pro-active" with regards to the upload/download side.

However, it should be noted that is ISN'T in the projects favour to dish out too many WUs, if a host doesn't have a half decent chance of crunching to the end of each one.

3) Past history of your efforts to crunch WUs for each project are stored on your PC...and of course on your account with each project....so it might be good to run down all your projects to zero, uninstall BOINC, delete the folder, use a good Registry Editor to delete any other references to BOINC and then install afresh.

But of course your "previous" history (for this same host PC) with each project is still held on your account, so if you use the SAME PC, then you might still be at square one

(You could "fool" each project by using the same PC, but under/over clocked very slightly to different CPU speed - then the project will think it's an entirely NEW host and hence your history will not be carried over).

regards,

Tim

[Rockinfroggi]

There are a few things that may be relevant:

1) - The connection via the ISP gets snarled up.
This can affect the upload and download of WUs and hence BOINC is left twiddling it's thumbs coz it cannot get to what it needs.

You can check the message logs, but they don't really tell you about the quality of the connection - so if something fails, there isn't a good report you can get which says why (and what you cna tehrefore do to fix it.

2) BOINC uses it's own "metrics" to decide when it needs more work.
Over time the core BOINC program has been changed a bit so that it can be a little more "pro-active" with regards to the upload/download side.

However, it should be noted that is ISN'T in the projects favour to dish out too many WUs, if a host doesn't have a half decent chance of crunching to the end of each one.

3) Past history of your efforts to crunch WUs for each project are stored on your PC...and of course on your account with each project....so it might be good to run down all your projects to zero, uninstall BOINC, delete the folder, use a good Registry Editor to delete any other references to BOINC and then install afresh.

But of course your "previous" history (for this same host PC) with each project is still held on your account, so if you use the SAME PC, then you might still be at square one

(You could "fool" each project by using the same PC, but under/over clocked very slightly to different CPU speed - then the project will think it's an entirely NEW host and hence your history will not be carried over).

regards,

Tim

Thanks Tim,

With regards to the internet connection, I did think it may be due to MJ12 taking up my bandwidth, but I've often put this on snooze for hours at a time with no visible improvement.

As far as fooling the projects if I were to delete and start again, would changing the host names do the same as changing the CPU speed?


Gary.

[MAVRIK]

I had the same problem untill i checked my prefs,it had been set to connect every 0.3 days,set this to ten and got hundreds of work units.I have no problems now.

[Rockinfroggi]

I had the same problem untill i checked my prefs,it had been set to connect every 0.3 days,set this to ten and got hundreds of work units.I have no problems now.

Mine was set to 2 days and I changed it to 3 days with no effect.


Gary.

[UBT - Timbo]

Thanks Tim,

With regards to the internet connection, I did think it may be due to MJ12 taking up my bandwidth, but I've often put this on snooze for hours at a time with no visible improvement.

As far as fooling the projects if I were to delete and start again, would changing the host names do the same as changing the CPU speed?


Gary.

TBH: I don't know if changing the host name might do the job.

IIRC, the important thing is the email addy you use....as that "generates" the CPID, which is then assigned to your hosts....so, maybe signing up with a different email will be enough....

BUT then your previous credits won't align with your NEW credits...!

regards,

Tim

[RodEllery]

The scheduler in the latest couple of manager releases leaves a lot to be desired if you have many projects attached. It seems to use a formula based on the the % of the project weightings assigned to each project.

In my case I have 30+ attached projects on one machine all with approx equal weightings so the sheduler seems to calculate that each project will get approx 3% of the available time. This means that it will download only the work that will be completed if the WUs get 3% of the time, despite the fact that there are only 1 or 2 projects with WUs already downloaded and running/ready to run so a new WU would get say 25-30% of the available CPU time.

Other machines with only 2-3 attached projects download loads of WUs at once and process normally.

I think if you want loads of WUs and a sizeable cache then the best bet id to detach from projects you don't want to immediately process and re=attach later.

<EDIT>
It's also crap at seeing work through to completion. At present I gave 2 jobs swapped out one with 3 mins to go, the other with 16 mins as it downloaded a new WU for another project with an earlier report date so i'ts processing that till it downloads another newer one. I regularly suspend network activity, suspend the running task and let all the others go through in order not to have 10 partially completed WUs on the go.
</EDIT>


On another subject:-

Anyone tried the 5.10 manager yet? Pirates are saying it will become mandatory in about 10 months time if you wish to crunch for them.

[Rockinfroggi]

Although it's a pain in my little bottom bits, I think I'll leave it as it is for the moment and see if the next update makes any difference.

Rod could have a point as it did seem to get worse after signing up to a number of projects in one hit.



Boincing Frog

[Ram Raider]

I'm using 5.10.0 and I've had no problems with it. In fact, as far as the scheduler is concerned, it's a vast improvement on previous versions and I wholeheartedly recommend it.

I'm using it in a variety of environments including one machine that is attached to every project going and it works very well without any need to micro-manage.

As Rod points out, it does take account of the resource share when requesting work but this is essential with so many projects as otherwise there would be a risk of missing deadlines. One difference (I think) from 5.8.16, is that the latest version will download additional work accordingly so that the cache is always full.

There are also other improvements so that projects are not pre-empted so often, especially when downloading work with very short deadlines eg Chess.

I've also found that changing the "switch between applications" setting to a higher value will keep the number of concurrent workunits in progress to a minimum. Mine is set to 180 minutes and this seems to work well as most projects have workunits that will complete in this time.

One final improvement, you can set the cache to zero and specify the work buffer using the new "maintain work for an additional" parameter. This means that workunits are usually reported immediately after they finish uploading.

So go on and try it - you may be pleasantly surprised.

Hope this helps. It's about time I reciprocated.

Richard

[RodEllery]

So go on and try it - you may be pleasantly surprised.

Ok giving it a whirl now. Have cache at 0 and additional work set to 10 days.

On one machine attached to 3 projects it requested work from malaria. This is what I got

07/06/2007 22:57:45|malariacontrol.net beta|Requesting 627494 seconds of new work
07/06/2007 22:57:50|malariacontrol.net beta|Scheduler RPC succeeded [server version 509]
07/06/2007 22:57:50|malariacontrol.net beta|Message from server: No work sent
07/06/2007 22:57:50|malariacontrol.net beta|Message from server: (won't finish in time) Computer on 81.0% of time, BOINC on 99.7% of that, this project gets 33.3% of that

at the time the machine had about 20 SETI wus (average about 7 hrs), 3 ABC average 1 1/2 hrs, and 1 malaria with an hour left. The deadlines for the SET are 25th of June, ABC 14th June, and Malaria 11th June. The machine is a has a dual core processor.  I fail to see why no malaria WUs were downloaded.

However once the malaria WU completed another (45 min) WU was downloaded imediately but only the one. This then started running immediately. Something ain't right somewhere but I am now thinking that the problen doesn't lie in the BOINC client but in some versions of the server software installed by some projects.

<edit>
It is now continually requesting work from malaria, backing off for longer and longer periods between each request, and getting the same response as before. (The other projects are set to 'no new work')
</edit>

[Rockinfroggi]

I'm using 5.10.0 and I've had no problems with it. In fact, as far as the scheduler is concerned, it's a vast improvement on previous versions and I wholeheartedly recommend it.

I'm using it in a variety of environments including one machine that is attached to every project going and it works very well without any need to micro-manage.

As Rod points out, it does take account of the resource share when requesting work but this is essential with so many projects as otherwise there would be a risk of missing deadlines. One difference (I think) from 5.8.16, is that the latest version will download additional work accordingly so that the cache is always full.

There are also other improvements so that projects are not pre-empted so often, especially when downloading work with very short deadlines eg Chess.

I've also found that changing the "switch between applications" setting to a higher value will keep the number of concurrent workunits in progress to a minimum. Mine is set to 180 minutes and this seems to work well as most projects have workunits that will complete in this time.

One final improvement, you can set the cache to zero and specify the work buffer using the new "maintain work for an additional" parameter. This means that workunits are usually reported immediately after they finish uploading.

So go on and try it - you may be pleasantly surprised.

Hope this helps. It's about time I reciprocated.

Richard

Thanks for that Richard, I'll try it out on one system later today and see how it runs.
I tend only to have one project at a time running on each system bar the odd Pirates or Belgian Beer.

I think I also need to look at detaching from some projects I rarely crunch  and things like Wep which Bam has attached me to on Windows clients but is Linux only.


Gary.

[UBT - Mikee]

07/06/2007 22:57:45|malariacontrol.net beta|Requesting 627494 seconds of new work
07/06/2007 22:57:50|malariacontrol.net beta|Scheduler RPC succeeded [server version 509]
07/06/2007 22:57:50|malariacontrol.net beta|Message from server: No work sent
07/06/2007 22:57:50|malariacontrol.net beta|Message from server: (won't finish in time) Computer on 81.0% of time, BOINC on 99.7% of that, this project gets 33.3% of that

I had that on the laptop except mine said it was 'Computer on 15%'. I left alone on for a couple of days and the message went away. That was versions 5.8.16 and version 5.10.0

[Ram Raider]

It looks like there is still some scope for improvement then!

I haven't seen that message on my environment but I have a fairly small cache of 6 hours. I find that works really well and I've never even come close to running out of work, but then I usually have work units from 20 or so projects waiting to run.

From what I have read, the overriding aim of the scheduler is not to miss deadlines so it's understandable that it takes a rather pessimistic approach, especially if you have a cache setting greather than the deadline of Malaria.

Another possible explanantion is that you have a high duration correction factor for Malaria so that the scheduler thinks the work units will take much longer than in reality.

If you use BoincView you can see this value in the projects tab otherwise you need to look inside client_state.xml

I use BoincView all the time, it's much better than the standard Boinc Manager and gives you much more information on what is going on.

Richard

[RodEllery]

Well things are now happenning. It downloaded about 40 Malaria WUs overnight and is now crunching them quite merrily (but not giving the ABC or Seti WUs a look-in).

On the machine with loads of projects attached it went a bit ballistic to start with. The first 10 or so projects it tried to download from had no work available. It then looped round thes projects requesting more and more work, occasionally trying a project further down the list. This seemed to be because these projects were backing off for only a minute or so before re-requesting work and the whole sycle started again. Eventually a couple of projects complained that the amount of work requested was greater than the maximum allowed. I shut down the client, cleared down the LTD and STD values to zero for all projects and restarted. This seems to have sorted the problem for now but I wonder what will happen when the projects with no work get back to the top of the LTD table.

Ho-Hum. Let's just wait and see.