Hi, I used the demo CD to test suspending a domain to disk, hoping to migrate suspended domains between Xen machines. The domain reported 62940 kB total memory in /proc/meminfo, with 29800 cached. This happily suspended to a 22965k gzip file, which expands to 65610k when decompressed. Am I right in thinking that the cache pages were be included, and if so are there any thoughts on removing them? I seem to recall that before swsusp actually starts its suspension of memory to disk scripts prepare the machine state and cache pages are not written out. This means that full performance isn''t returned immediately after resumption, but presumably it''s just as fast to re-read the cache back from files on disk rather than pages from a suspend file, and it minimises machine state. Cheers, Sean. -- Sean Atkinson <sean@netproject.com> Netproject ------------------------------------------------------- This SF.Net email is sponsored by: IBM Linux Tutorials Free Linux tutorial presented by Daniel Robbins, President and CEO of GenToo technologies. Learn everything from fundamentals to system administration.http://ads.osdn.com/?ad_id=1470&alloc_id=3638&op=click _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
> Hi, > > I used the demo CD to test suspending a domain to disk, hoping to > migrate suspended domains between Xen machines. The domain reported > 62940 kB total memory in /proc/meminfo, with 29800 cached. This happily > suspended to a 22965k gzip file, which expands to 65610k when > decompressed. > > Am I right in thinking that the cache pages were be included, and if so > are there any thoughts on removing them?You could in principle use the balloon driver to minimise the state before doing the suspend. I fear the balloon driver has rotted slightly as we don''t use it much. We have plans to revive and improve it, but don''t have the man power. I''d be happy to explain to someone what needs doing if we have a volunteer.> I seem to recall that before swsusp actually starts its suspension of > memory to disk scripts prepare the machine state and cache pages are not > written out. This means that full performance isn''t returned > immediately after resumption, but presumably it''s just as fast to > re-read the cache back from files on disk rather than pages from a > suspend file, and it minimises machine state.I would imagine that restoring the buffer cache in one go is typically faster than taking a whole string of faults, but it depends on how well the buffer cache is working for the application. Best, Ian ------------------------------------------------------- This SF.Net email is sponsored by: IBM Linux Tutorials Free Linux tutorial presented by Daniel Robbins, President and CEO of GenToo technologies. Learn everything from fundamentals to system administration.http://ads.osdn.com/?ad_id=1470&alloc_id=3638&op=click _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
[...] >> I seem to recall that before swsusp actually starts its suspension >> of memory to disk scripts prepare the machine state and cache pages >> are not written out. This means that full performance isn''t >> returned immediately after resumption, but presumably it''s just as >> fast to re-read the cache back from files on disk rather than pages >> from a suspend file, and it minimises machine state.>>>>> "Ian" == Ian Pratt <Ian.Pratt@cl.cam.ac.uk> writes:Ian> I would imagine that restoring the buffer cache in one go is Ian> typically faster than taking a whole string of faults, but it Ian> depends on how well the buffer cache is working for the Ian> application. It is definitely faster on a laptop. Swsusp used to discard (actually "eat") cache, but it doesn''t do that anymore. It is much better to preserve it across suspend/resume cycles, otherwise you end up with a flurry of disk activity right after you resume. On a laptop it also helps, because if you preserve it, you allow the disk to spin down. If you discard it every time, there is a much bigger chance that the data you need has to be read from the hard drive. For those who care about image size and suspend speed, swsusp2 now includes LZF compression. I get average savings of about 40% and increased suspend speed. --J.