Are each of your 4 servers ALSO metadata servers? (wasn't sure if you had told me that or not) Just in case, you should have all servers defined as I/O and Metadata servers.
The response time on a typical "ls" is affected by the network congestion and the load on the servers. Thus, "ls" will perform more slowly when congestion and loads are high than when they are low. In addition, "ls" also sorts all the data before it is displayed, which causes the perception that OrangeFS is slow, especially if there are many entries in the directory. Try this: when "ls" is performing slowly, use pvfs2-ls and you will see the difference between going through the kernel to access OrangeFS and direct access to OrangeFS! Or, try using "ls -U", which tells "ls" not to sort. In a near-term release, we will have a user library that will allow commands like "ls" to run more efficiently against OrangeFS. Keep your eyes open for that! Becky On Wed, Jan 11, 2012 at 2:54 PM, Yves Revaz <[email protected]> wrote: > ** > On 01/11/2012 07:59 PM, Becky Ligon wrote: > > The information for the mount command is only used once when the client > starts up to know where to get the config file. So, unless you have > hundreds or thousands of clients that you start at once, it really doesn't > matter. If you are in a situation where you have lots of nodes starting > the client simultaneously, then balancing of nodes per server is a good > idea, just as you have described. > > ok, perfect, thanks. > > > I'm not following your question about disk space. Please give me more > information. > > > ok, I will try to by more clear. > > Those last two days, my pvfs2 file system seems to be very very slow. It > takes a lot of time > to copy small files and a simple ls needs time to return an answer. > > Looking at the file system, I see that it is nearly full (95%), so I was > wondering if the > fact that all disks participating to the file system are nearly full could > impact on the speed of the file system itself. > I think this occurs with other file systems, like zfs for example. > > However, interestingly just now it works much better, while it is still > full at a 95% level. So I'm trying > to find another explanation. In fact, two days ago, two among four of the > deamon server (pvfs2-server) crashes. > I restarted them. Do you think they could be some link between the > slowness of the file system and these > crashes ? > > Thanks, > > yves > > > > > > > > > Becky > > On Wed, Jan 11, 2012 at 11:29 AM, Yves Revaz <[email protected]> wrote: > >> Dear List, >> >> I'm using pvfs2 successfully on a cluster since now about 2-3 month, >> with only one or two minor problems. >> >> However, I have two questions to ask: >> >> 1) I have 4 masters that provides disk (all are meta data servers), say : >> >> m001,m002,m003,m004 >> >> and n slaves that must access the servers. >> pvfs2 is mounted using the kernel module. >> Actually, for each slave, I use the following command : >> >> mount -t pvfs2 ib://m001:3335/pvfs2-fs /SCRATCH >> >> Is it a problem that all slaves goes through m001 ? Maybe it should >> be better that 1/4 of the slave uses m001, 1/4 m002 , etc >> in order to avoid a bottleneck, no ? I have no idea how this works. >> >> 2) Those last days, the pvfs2 disk space seems very low. Can it be simply >> due to >> the fact that all disks are nearly full (actually 95% )? >> >> Filesystem Size Used Avail Use% Mounted on >> 3.4T 3.3T 189G 95% /SCRATCH >> >> >> >> With best regards, >> >> >> yves >> >> >> >> >> -- >> (o o) >> --------------------------------------------oOO--(_)--OOo------- >> Dr. Yves Revaz >> Laboratory of Astrophysics >> Ecole Polytechnique Fédérale de Lausanne (EPFL) >> Observatoire de Sauverny Tel : ++ 41 22 379 24 28 >> 51. Ch. des Maillettes Fax : ++ 41 22 379 22 05 >> 1290 Sauverny e-mail : [email protected] >> SWITZERLAND Web : http://www.lunix.ch/revaz/ >> ---------------------------------------------------------------- >> >> _______________________________________________ >> 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 > > > > > -- > (o o) > --------------------------------------------oOO--(_)--OOo------- > Yves Revaz > Laboratory of Astrophysics EPFL > Observatoire de Sauverny Tel : ++ 41 22 379 24 28 > > 51. Ch. des Maillettes Fax : ++ 41 22 379 22 05 > 1290 Sauverny e-mail : [email protected] > SWITZERLAND Web : http://www.lunix.ch/revaz/ > ---------------------------------------------------------------- > > -- 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
