Andrew: Right now, we don't have special files on our development list, but that could change if there is enough demand for it. No promises!
And, yes, you will get better throughput by not syncing meta or data files and not using data=writeback; however, if you're in a production environment, you don't want to do that, unless you don't care about any kind of recovery. OrangeFS doesn't provide any kind of transaction processing or data recovery; it MUST be provided by the environment. (FYI: we ARE developing redundancy into the system now.) For example, in our production environment at Clemson, we have meta data sync turned on, with data=ordered turned on the ext4 system underneath, and we have data sync turned off with data=writeback turned on the underlying ext4 filesystem. We also have the data in a RAID-5 and the meta data mirrored. Becky On Tue, Mar 13, 2012 at 2:53 PM, Andrew Savchenko <[email protected]> wrote: > Hello, > > is it possible to create a fifo and a socket special files on the > pvfs2 file system? On mounted orangefs-2.8.5 both mkfifo and bind > calls return errno 95: Operation not supported. > > As far as I understand, these features are not implemented currently. > Are there any plans to do this or support of special files does not > fit into pvfs2 purposes and goals? (Something tells me the latter > assumption in the correct one: pvfs2 is not a general file system > after all.) > > ***************************************************************** > > Why I need this at all? I tried one crazy experiment: to use pvfs2 > as a /home directory. Yes, I know, you wouldn't approve this approach. > But otherwise I need to spare one of storage devices for /home and to > teach users that they need to use special /mnt/pvfs2 directory for > their main storage (this is the worst problem). > > Pvfs2 performance proved to be excellent on large files, large i/o > chunks and (as expected and well documented) is rather poor on a > bunch of small files (just untar kernel sources to see what I mean). > > And I found one interesting thing for small files over pvfs2 over > ext4 over sata drive: while > TroveSyncMeta no > definitely helps by about 30% of speed improvement (and > TroveSyncData is disabled by default config), the greatest > performance benefit is from disabled journal on underlying ext4. > It gave me 10 (ten!) times improved throughoutput compared to > data=writeback. > > Of course this comes at the high risk of data corruption if no > additional measures are taken. Maybe this observation will be useful > for someone. > > Best regards, > Andrew Savchenko > > _______________________________________________ > Pvfs2-users mailing list > [email protected] > http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users > > -- Becky Ligon OrangeFS Support and Development Omnibond Systems Anderson, South Carolina
_______________________________________________ Pvfs2-users mailing list [email protected] http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
