Corrado Zoccolo wrote: > Hi > I have few suggestions for sreadahead, I hope this list is the proper > place for them. > > * earlied versions of sreadahead used to put files under /etc, while > version 1.0 uses /var. I find /etc to be more convenient, since I use > a separate /var partition, and it is not yet mounted when > rc.early.local is invoked (i.e. where I would like to start > sreadahead)
sreadahead uses /var to mount debugfs at the early stage, hence my choice to move this location to /var (I think /etc should be clear of mounts). I also think that the sreadahead pack file is not a configuration file, and as such doesn't belong in /etc either. IMHO /var is the right location, if you want to have a specific /var partition, then I suggest that you delay starting up sreadahead until /var is mounted. > * use posix_fadvise(fd,off,len,POSIX_FADV_WILLNEED) instead of > readahead(fs,off,len): posix_fadvise is non-blocking until there is > room in the queue, so you can get rid of multithreading, and it will > better interact with the block layer, allowing it to reorder more > requests according to disk placement. I'm currently contemplating removing all the threading support, because I'm not seeing an improvement - even on SSD's. > * export a plain file list of the files that are opened during boot. > this list can be used by external script to optimize on-disk file > placement for rotational disks. would a simple `dump` utility do? > * give an option to skip readahead of large contiguous blocks. This > improves rotational disk performance when the disk is the bottleneck, > since we prefetch all random accesses, that normally incur in high > latency, and let the processes complete their sequential reads, that > are faster, only when needed, with less contention. obviously for spinning discs only. I haven't really optimized sreadahead yet for this case, but this sounds like an ok idea. Have you tried it out? Auke _______________________________________________ Power mailing list [email protected] http://www.bughost.org/mailman/listinfo/power
