>T. Horsnell wrote: > >> 1. Is there a way to restrict the amount of >> memory consumed by the UBC? >> >> 2. Is there any way of mounting a filesystem >> such that its data doesnt pass through >> the UBC at all? > >There is no consumption of memory. If the buffering system was not >using it the memory would just be sitting there unusued at all. All the >buffer memory is ready for allocation, just requiring flush if dirty. Yes, but when another application starts up, and needs memory which has been consumed in UBC, *it* has to wait until sufficient buffer has been flushed to disk. And since there is no point in buffering the data in the first place if its not going to be re-accessed in a short time, that delay could possibly be avoided. > >Similarly the "substantial delay" for syncing is not real either, the >total time for an (app dumping into the buffer + flushing time) should >be the same as an unbuffered app dumping direct. Its not the speed of the app writing to disk (via the UBC) that I'm talking about. Its the startup time of other apps which need memory which can only be acquired by flushing the cache. If I write a 20GB file, effectively all my 'spare' memory gets used for cache. If I then want to run a new application at the same time, I have to wait while sufficient cache is flushed. And the writing of that 20GB file may not have benefited at all by being cached. Cheers, Terry. > >-Andy >-- >fedora-list mailing list >fedora-list@xxxxxxxxxx >To unsubscribe: https://www.redhat.com/mailman/listinfo/fedora-list
