Bill Davidsen wrote: > Roberto Ragusa wrote: >> The decision to not save buffers and cached is debatable. Even if >> it is memory which can be read again from the disk, it is MUCH >> faster to read from the swap image in a contiguous fashion, than >> to seek everywhere for minutes after a suspend. > > That argument makes no sense for buffers, you don't want to compress > them, write them to swap, read them in, decompress them, all so you can > then write them to the filesystem anyway. Buffers are not always dirty data. After a "sync" command you don't see them drop to zero. Even if they were dirty data, it can make sense to store them sequentially and compressed, instead of writing them in the right places (which could involve many many seeks). I'm perfectly aware that they will have to be properly written after resuming, but if I say "hibernate" I want the machine to hibernate as soon as possible, not to force the execution of any pending activity. I would run "sync" before hibernation, if I wanted to write dirty buffers (in any case, suspending scripts often do "sync" themselves before starting). >> That is one of the really good things which only tuxonice has. > > I hope you're remembering that wrong, if it's going to be written to > disk there's no benefit to doing a lot now so you can slow the restore > and then do all that i/o anyway. You can make the argument for cache, at > least you might save so i/o after resume. I rechecked on the net just now, and tuxonice does indeed save buffers and cached by default. >> I hope to see it finally merged. > > Don't hold your breath, for years I kept pointing out issue with suspend > and being told "stop whining and send patches," and I would say "the > patch is suspend2." It's been renamed, but still mostly not merged, and > new bugs have been added. I used suspend2/tuxonice a few years ago and found it really good. The official suspend stuff never really worked. Nowadays I'm only using suspend-to-ram (damned Nvidia proprietary crap). >>> so it's really hard to guess how much >>> swap will actually be used. >> >> But it is really simple to decide the size of the swap >> partition: use a big one. >> There is really no disadvantage to have a bigger one. >> >> 2 GB RAM? -> 5 GB swap >> >> more than enough to have 1gb swap used and suspend everything; >> probably large enough even when you upgrade to 4 GB RAM. >> >> I don't think anyone really cares about 2 or 3 GB of disk >> on a modern machine (having 2 GB of RAM makes it modern). >> > Haven't done much embedded work? Think 512M RAM, 2GB non-volatile > storage, maybe another bit of EAPROM for firmware. That's not Fedora, > but it definitely is suspend/resume. But this thread was about someone with a 64 bit machine and a new 1TB disk. 2GB "wasted" on a 1000GB disk is not a big issue. When you run Linux on a toaster, you decide differently. :-) -- Roberto Ragusa mail at robertoragusa.it -- fedora-list mailing list fedora-list@xxxxxxxxxx To unsubscribe: https://www.redhat.com/mailman/listinfo/fedora-list Guidelines: http://fedoraproject.org/wiki/Communicate/MailingListGuidelines
