I noticed a terrible slow-down on my P-4 3.0Ghz 1GB computer and traced it to two instances of Einstein executables named einstein_S5GC1_3.02_windows_intelx86__S5GCESSE2.exe which were doing I/O to a hard disk. The total space used in the Einstein folder is 617MB, all of which could be read in about 160,000 reads of 4K bytes each. The actual I/O count when they went into "crunch" mode is about 12,700,000.
Please re-visit the algorithms being used and reduce this access to a reasonable number of I/Os.
Alternatively, don't send more than ONE of these types of WUs to any one computer at a time. (One plays somewhat nicely; two fight with each other for the I/O.)
Thanks in advance.
Copyright © 2024 Einstein@Home. All rights reserved.
Improved disk I/O requested
)
Thanks for the hint. I'll look into the reading routine when I find some time to work on the app code again.
BM
BM
Yes I concur with what looks
)
Yes I concur with what looks like excessive writes and this is even with lots of free physical memory available. Mind you this does give me a reason to buy an SSD for 'experimental' reasons.
RE: Yes I concur with what
)
AFAIU Darrell was talking about the reading of the data at App start, before the actual computation begins.
If you see 'excessive writes' during computation, the only reason for this could be checkpointing. Try reducing the "write to disk at most every xxx seconds" setting in your computation preferences.
BM
BM
I'd be interested to know if
)
I'd be interested to know if the new generation of S5GC1HF Apps brought some improvement regarding app startup / reading data for you. Please report at this thread.
BM
BM