S5R5 making no progress on Windows 2000 Standard Server
23 Jan 2009 17:53:12 UTC
Topic 194154
(moderation:
Can anyone come up with any ideas for this one? (Previously posted in Number Crunching, where we eliminated Hyperthreading - there isn't any)
Quote:
I'm having problems getting one machine to run the S5R5 application. It's host 475735, my Windows 2000 standard server (SP4). It's just my domestic file/print server, not a domain controller or anything. It's been running earlier versions of Einstein just fine (see the host join date/credit), and it's continuing to run SETI without problems. BOINC is v5.10.13 installed as a service - no recent change.
The problem with S5R5 is that tasks (well, the only S5R5 task it's been assigned so far) starts to run, but makes no progress at all. I was away at the weekend, and the app ran for well over a day with still 0.000% progress showing.
Also, once the app starts, I can't find any way of stopping it. If I suspend the task via BOINCManager or BoincView, it continues to run at 99% CPU utilisation. Likewise if I shut down the BOINC service. I can't even kill the Einstein process with Task Manager - it tells me 'access denied'. The only way I can get back to productive work (e.g. on SETI) is to reboot the whole computer.
The CPU is a single-core P4 Northwood, with 512MB RAM. It's a very close match to my host 1036916, which runs S5R5 with no problems under XP SP3. Any ideas?
S5R5 making no progress on Windows 2000 Standard Server
)
He can't stop it because he installed as a service. I have the same problem on a Vista machine. My wife kept stopping Boinc, so I installed it as a service and now she can't do it anymore.
As for why it isn't moving forward, I can't say. But installing a newer version might work better, if it is Win 2k compatible.
RE: ... The problem with
)
I've never seen an instance where the app cannot be killed but I have seen plenty of cases where the app gets stuck at a certain percentage and makes no further progress, no matter how long it continues to run. In my cases (which have been occurring for some time - before the start of R4) it is always possible to kick start things just by stopping and restarting BOINC. Your situation is obviously different and I can't really offer any useful suggestions.
Because I can't really closely monitor all my hosts, the situation can go undetected for days or even weeks. I have found that it is good to monitor both time of last contact and CPU efficiency on the website. If crunching halts while the machine seems otherwise alive and well, contact tends to stop and CPU efficiency can quickly take a dive. Each day I tend to have a look at any hosts at the bottom of the hosts list and then look at CPU efficiency of any that seem unusually late in making contact.
I'm sorry I can't really offer anything useful. I'd be interested to know what it finally turns out to be if you eventually get it going again. I would have thought there would be quite a few more reports if the app had a habit of doing this in general.
Good luck with sorting it out!
Cheers,
Gary.
RE: He can't stop it
)
That's not correct. It is BOINC that is installed as a service and he can stop BOINC any time he wants. For some reason it's the science app that is unkillable but that has nothing to do with BOINC running as a service.
I hope your wife isn't reading these boards :-). She could easily kill the BOINC service any time she wants to, even if she doesn't already have sufficient priviledges. She just has to beat you up and extract the Administrator password ... :-).
There is no newer version of the program that is unkillable. The R5 science app (just released) is the latest there is. He could upgrade to a later version of BOINC but it's not clear whether that would make any difference.
Cheers,
Gary.
RE: RE: ... installing a
)
Perhaps he meant I should install Windows Server 2008 - I don't think I'll bother, on that hardware.
Gary is right: I can suspend any task, or even whole projects, via BoincView. This S5R5 shows 'suspended' OK, but still eats 99% CPU - the alternate project task shows as 'running', but gets a minute number of CPU cycles.
I can also shut down the whole of BOINC (hint for Mikey's wife: services.msc), but this task just keeps on running like that pesky bunny....
I'm giving it another try now, but still zero progress after 20 minutes. I'll give it another 10, then try aborting, cleaning client_state and the project directory, resetting the project, and doing a clean re-fetch of both apps and data.
One small irritation with the app and W2K: Task Manager truncates the app name at the first period/full stop/point, so all I see is einstein_S5R5_3 - I can't tell which SSE version the switcher has chosen.
RE: RE: RE: ...
)
I actually meant a newer version of Boinc that was still Win2k compatible.
She does not read the forums but I will try that and see if it stops the crunching. I can exit Boinc go into Task Manager and see 2 instances still happily crunching, it is a dual core machine. If I then restart Boinc, then exit again, no changes. Still crunching along. I was trying to stop it so I could do a new anti-virus install the other day. I was switching to a new program and don't like anything else running when I install new programs. In the end it was okay, I was just a frustrated camper for a bit.
RE: She does not read the
)
When you say 'exit BOINC', I think you are actually just exiting the BOINC Manager - not the actual BOINC worker (daemon) application. Watch Task Manager while you do it: I think you'll see boincmgr.exe disappear from the processes list, but boinc.exe stay there.
You can shut down the worker thread in many ways: type 'NET STOP BOINC' at a command prompt, click 'Shut down connected client...' in BOINC Manager (advanced view, advanced menu). But the service control panel (run services.msc, or drill down through Control Panel to Administrative Tools) is my favourite, because it works the same on all versions of Windows (2K and up) and all versions of BOINC, and it allows you to see the state of the service while you're controlling it.
While I was composing that
)
While I was composing that reply to Mikey, BOINC/Einstein was getting its act together. The thorough purge and re-download seems to have done the trick: I've made some progress on one task, suspended it, and made progress on a second. So the finger of suspicion points at a corrupt download, rather than an incompatibility with W2KSS. Sorry for the false alarm.
(and about that unkillable application: I found I had the service running under the local system account. I've now re-configured it to run under my own login: when I get a chance, I'll check if that makes it killable).
RE: ... I can exit Boinc go
)
As Richard points out, you need to be careful to differentiate between BOINC (boinc.exe) and BOINC Manager (boincmgr.exe) the gui interface to BOINC. Also remember this - fully exiting BOINC Manager will only stop BOINC if BOINC Manager started BOINC in the first place. With a single user install, it is BOINC Manager that starts BOINC. With a service install, BOINC will normally be running (started by the system) before you launch BOINC Manager. The Manager simply finds and attaches to the already running BOINC. When you exit the Manager, BOINC is not affected and continues to run in the background - which is what you were seeing.
I find it very useful to create two scripts in the BOINC main folder and call them StartBoinc.cmd and StopBoinc.cmd. They contain just the one line "net start boinc" and "net stop boinc" respectively. After you create and save the scripts just rightclick each one and select "send to desktop (create shortcut)" and you will have two icons on your desktop immediately available to start and stop the BOINC service whenever you need to.
The slightly more complicated way is to rightclick the "My Computer" icon and select "Manage" from the context menu. This throws up a management console window from which you can do many things. If you expand the "Services & Apps" entry in the left pane and select "Services" you will get a full list of all your services in the right pane, with the option of doing several things with them, including starting and stopping or even changing their properties. This is exactly what you get if you had run services.msc as Richard mentions. Experiment with that and see how convenient it is.
Hope some of this helps.
Cheers,
Gary.
RE: While I was composing
)
No problem! That's very good news that you have the app actually making progress.
I'm sure that will will be the answer since the app, when started, would have previously been owned by the local system account and you, as a mere mortal, would have been totally ignored in your attempts to kill something you didn't own :-).
When you do a service install, the installer asks you for the account name and password under which BOINC will run. If your user account doesn't have a password, BOINC wont accept it and one of the alternatives is to run under the local system account instead. Maybe this is how it got to be the local system account in the first place.
Cheers,
Gary.
RE: When you do a service
)
Most of my BOINC installations started as BOINC v5, and I long ago settled on the formula that suits my home network (which is adequately secure behing a good firewall router - business network administrators wouldn't be happy with all of this).
1) Use the same user name and password on all machines
2) Use AutoAdminLogon so the machines start up without prompting for a password
3) Install BOINC as a service using my standard logon credentials
4) Once installed, switch the service to local system account with desktop interaction. In theory, that meant I could see the v5 graphics, but I very rarely bothered.
The benefit of the consistent username/password is that I can use network shares to move files from machine to machine, from Windows 98 to XP Home to Vista, without having to pay for Pro licenses or a domain controller.
Anyway, back on topic - the server has reported and validated task 115557335, so it's back in business. Thanks Gary.