But why stop at version 5.8? Why not make people go all the way to 6.x instead?
My observation with 6.x is, that it fails to work in a more complex environment
than a single computer.
You need an own home directory on any computer to run it on more than one
computer of a linux-cluster, that shares accounts and home directories.
In such a construction you can run one computer with 6.x, the other with 5.x
or stop BOINC on them - at least I found no other simple solution without
generating superfluous additional accounts. If someone knows, how to optimise
6.x to use only information in an arbitrary subdirectory, this would be an
option for me to update BOINC from 5.10 to 6.x. Luckily there seems to be no
urgent problem with 5.10 ;o)
I'm pleased to note that Bernd has implemented some improvements in the Einstein server scheduler.
With my BOINC v5.10 installation (highest permitted on a Windows Domain Controller), the new scheduler has issued the SSE2 application, instead of the plain-vanilla app I was running before. This has cut down the running time from ~8 hours with the plain vanilla Global Correlations S5 HF search application, to ~5.5 hours with the SSE2 optimised version.
If you are already running S5GC1HF under BOINC v5.10 or earlier, you won't get the optimised version automatically until the next version upgrade on the server. But if you perform a project reset, you should get it straight away. Note that resetting the project discards all work in progress or ready to report: only do it at a time when you have minimal work on hand, or after you have drained your cache by setting 'No New Tasks'.
If someone knows, how to optimise
6.x to use only information in an arbitrary subdirectory, this would be an
option for me to update BOINC from 5.10 to 6.x
There are two command-line options of the client that might help:
* "--insecure" should switch off the "sandboxing" (requiring a "boinc" user account)
* "--dir /var/tmp/BOINC" allows to put the BOINC data directory anywhere you like, in the example to /var/tmp/BOINC, which should be local on all machines on a cluster (as an Admin I'd prefer /var/lib/boinc)
If you are already running S5GC1HF under BOINC v5.10 or earlier, you won't get the optimised version automatically until the next version upgrade on the server. But if you perform a project reset, you should get it straight away...
I did get the SSE2 version without resetting the project. :-D
I didn't even notice the change until I'd read your post. :-)
However, my Einstein cache had been empty before the download of einstein_S5GC1HF_3.06_windows_intelx86__S5GCESSE2.exe took place (due to debt issues with SETI I reckon).
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
If you are already running S5GC1HF under BOINC v5.10 or earlier, you won't get the optimised version automatically until the next version upgrade on the server. But if you perform a project reset, you should get it straight away...
I did get the SSE2 version without resetting the project. :-D
I didn't even notice the change until I'd read your post. :-)
However, my Einstein cache had been empty before the download of einstein_S5GC1HF_3.06_windows_intelx86__S5GCESSE2.exe took place (due to debt issues with SETI I reckon).
Gruß,
Gundolf
If you haven't downloaded any new work since the x.06 applications were installed on 8 December, yes, you'll get the SSE2 (or whichever version is appropriate) automatically with your next download. But if you have downloaded work since that date, you'll already have an x.06 app, but you won't get upgraded to SSE/SSE2 without a reset or until a new version is installed on the server.
I did check the release date of the application but only assumed I would have downloaded some work between then (12/8) and the download of the SSE2 application (12/11). However, my host seems to be slower than I've thought: the latest previous download had occured on 12/5.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
AFAIK, version 6 still does not run on a Windows domain controller. Until this is fixed, version 5 should still be supported!
Actually, BOINC v6 cannot be installed on a domain controller - it's the installation process with the normal Berkeley installer which fails: there are reports that it runs OK if you install it manually.
But quite a bit of this and the related threads have been about that problem: and Bernd has stated that he has no intention of raising the minimum BOINC version beyond v5.8, and has put considerable effort to ensure that BOINC v5 clients run as efficiently as possible.
So I trust this user will be told how to upgrade, since I don't know what the hell you're talking about. Such mysterious and clandestine motives. For most of us, we're just running your Boinc software. We don't know what the hell you're talking about. Talk to the majority of your users, not the geek specialists. Most of us are out here just trying to help out. Get a grip and give the illiterate Boinc crowd some slack.
RE: But why stop at
)
My observation with 6.x is, that it fails to work in a more complex environment
than a single computer.
You need an own home directory on any computer to run it on more than one
computer of a linux-cluster, that shares accounts and home directories.
In such a construction you can run one computer with 6.x, the other with 5.x
or stop BOINC on them - at least I found no other simple solution without
generating superfluous additional accounts. If someone knows, how to optimise
6.x to use only information in an arbitrary subdirectory, this would be an
option for me to update BOINC from 5.10 to 6.x. Luckily there seems to be no
urgent problem with 5.10 ;o)
Doesn't running the Client
)
Doesn't running the Client without sandboxing (i.e. with command-line option --insecure) help?
BM
BM
I'm pleased to note that
)
I'm pleased to note that Bernd has implemented some improvements in the Einstein server scheduler.
With my BOINC v5.10 installation (highest permitted on a Windows Domain Controller), the new scheduler has issued the SSE2 application, instead of the plain-vanilla app I was running before. This has cut down the running time from ~8 hours with the plain vanilla Global Correlations S5 HF search application, to ~5.5 hours with the SSE2 optimised version.
If you are already running S5GC1HF under BOINC v5.10 or earlier, you won't get the optimised version automatically until the next version upgrade on the server. But if you perform a project reset, you should get it straight away. Note that resetting the project discards all work in progress or ready to report: only do it at a time when you have minimal work on hand, or after you have drained your cache by setting 'No New Tasks'.
RE: If someone knows, how
)
There are two command-line options of the client that might help:
* "--insecure" should switch off the "sandboxing" (requiring a "boinc" user account)
* "--dir /var/tmp/BOINC" allows to put the BOINC data directory anywhere you like, in the example to /var/tmp/BOINC, which should be local on all machines on a cluster (as an Admin I'd prefer /var/lib/boinc)
BM
BM
RE: If you are already
)
I did get the SSE2 version without resetting the project. :-D
I didn't even notice the change until I'd read your post. :-)
However, my Einstein cache had been empty before the download of einstein_S5GC1HF_3.06_windows_intelx86__S5GCESSE2.exe took place (due to debt issues with SETI I reckon).
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
RE: RE: If you are
)
If you haven't downloaded any new work since the x.06 applications were installed on 8 December, yes, you'll get the SSE2 (or whichever version is appropriate) automatically with your next download. But if you have downloaded work since that date, you'll already have an x.06 app, but you won't get upgraded to SSE/SSE2 without a reset or until a new version is installed on the server.
Oops, I stand corrected. I
)
Oops, I stand corrected.
I did check the release date of the application but only assumed I would have downloaded some work between then (12/8) and the download of the SSE2 application (12/11). However, my host seems to be slower than I've thought: the latest previous download had occured on 12/5.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
AFAIK, version 6 still does
)
AFAIK, version 6 still does not run on a Windows domain controller. Until this is fixed, version 5 should still be supported!
JM
RE: AFAIK, version 6 still
)
Actually, BOINC v6 cannot be installed on a domain controller - it's the installation process with the normal Berkeley installer which fails: there are reports that it runs OK if you install it manually.
But quite a bit of this and the related threads have been about that problem: and Bernd has stated that he has no intention of raising the minimum BOINC version beyond v5.8, and has put considerable effort to ensure that BOINC v5 clients run as efficiently as possible.
So I trust this user will be
)
So I trust this user will be told how to upgrade, since I don't know what the hell you're talking about. Such mysterious and clandestine motives. For most of us, we're just running your Boinc software. We don't know what the hell you're talking about. Talk to the majority of your users, not the geek specialists. Most of us are out here just trying to help out. Get a grip and give the illiterate Boinc crowd some slack.