Disk write question
11 Apr 2006 15:35:28 UTC
Topic 191071
(moderation:
I would like it that when BOINC was suspended that it would force a write to the disk. That way one wouldn't lose any work due to 'Write to disk at most every'. To clarify, I have it set to 900/s and if I have to restart my computer then I lose out on X amount of work progress (even if I suspend the project[s]). I'm not about forcing when a computer crashes or such.
I increased the disk write from default because I had read in the BOINC wiki that doing so would increase performance. To quote, "The last setting is the one for the 'write to disk interval'. This setting, made high enough can get you a lot more if you do not mind the risk of losing an hour or so of work." I am curious as to why, to save on cycle usage? Or less stress on the disk? All in all it is a minor irration but I'm curious.
Another question I have is how high is "high enough"? I have two computers working on Einstein (P4 3.0 1014MB & Cel. 2.4 254MB).
I had asked these questions on the Boinc message board and was told "Unfortunately this is likely to be impossible or impractical. Some science applications can not checkpoint on command due to the calculations involved" and "I do not know about science apps which perform better with longer 'Write to disk intervals'. For some they are hardcoded (climateprediction for example) for others they vary according to the WU (Rosetta for example). Longer 'write to disk' periods can reduce hard disk stress but that is hardly of any importance."
So, is "write disk interval" meaningless to Einstein? I suppose that is the real question which all others are dependent on.
Disk write question
)
Nevermorestr,
I don't think that opening up the checkpointing (write to disk..) interval has a noticable impact on performance in Einstein. This is done on the fly, and requires very little CPU usage. Modern harddrive mttf (mean time to failure) ratings are so high (100s of 1000s of hours) that stress need not be a consideration, either.
microcraft
"The arc of history is long, but it bends toward justice" - MLK
Thank you Michael for the
)
Thank you Michael for the useful and timely answer. I hope you are feeling well and good luck in your "battle".
RE: Thank you Michael for
)
nevermorestr,
I am grateful for your kind wishes. I think, that as the battle draws on, prayers and roots likely will play roles as large as the $800/day medications, and that the intentions of the most-high power for my future, addressible by petition, are the ultimate factor. All kind wishes and prayers and encouraging words replenish my energy when it ebbs, and I dare not give less than my utmost when many have given their best to me.
microcraft
"The arc of history is long, but it bends toward justice" - MLK
RE: I dare not give less
)
Too right mate! :-)
Otherwise, for example, I will be particularly p***ed off ......! :-)
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Set "Remove App from memory
)
Set "Remove App from memory when suspended" in your preferences. This would terminate the App when BOINC gets suspended, forcing a writing of the checkpoint.
BM
BM
RE: Set "Remove App from
)
Yes, terminate, but does not write a checkpoint. It restarts from last checkpoint written, which could be a while ago. Example is Climate and may never progress if done this way.
Lets talk about harddisk
)
Lets talk about harddisk stress.
I worked in a little software company as hardware technican.
One time a big customer wanted a software cange to write every change that was made at the terminal immediately to the harddisk.
Not long after the softwareupdate (about a week or so) the harddisks of the customers computers were breaking down, one after the other.
I suspect the constant movement of the read/write head was too much for the old mechanic of the harddiscs.
The disks had to be replaced and the software was being downgraded.
That was the first time i heared from the possibillity to downgrade.
BOINC itself is being
)
BOINC itself is being updated. This is supposed to make some pretty big changes in how often it saves, etc. So, when the new version comes out, we should have "safer" client_state.xml writes, and less disk writes.
RE: BOINC itself is being
)
Does anyone know if this was implemented? Or if not, why?
RE: Does anyone know if
)
The new version is on the "All versions" page, but they warn you it may be unstable. I believe this version does have some backup of the different xml files needed during the work. I personally have not tried it, but under the version details it says "Revamped logging solution." So hopefully it is.
Maybe some of the Beta testers can tell us if the fix has been implemented.