I don't have that many interfaces, but I do have almost 11,000 rrd files I upgraded to rrdtool 1.2.999, i didn't fancy going all the way to 1.3, 1.2.99 should still have all the extra performance patches, but didn't seem to make too much difference.
I did consider changing the mount options on the block device (noatime etc.), but this would have been a last resort as I would have had to remount the disk, I decided against this once the block level read-ahead fixed my problems. Reading through the rrdtool documentation it seems to blame the linux filesystem more than anything else, but I noticed a definite difference between 1.0.x and 1.2.x. Embarrassing spec: Dual P III @1.4GHz 256 MB RAM, 2 x 7200 SCSI Disk (RAID 1) 46 hosts, 1242 interfaces mostly all ethernet) still using old poller (against my own advice!). Cheers Martin LIMA David wrote: > I confirm that i had a problem with rrdtool but with lots of interfaces > (more than 10 000). > > I solved my problems using: > - a dedicated files system and block dev for RRD files > - upgrading rrd to 1.3 (still in beta) : Using Memory Mapped IO for updates > and severall performance patch > - reducing block level read-a-head: blockdev --setra 16 /dev/sda > - files system optimization: nodiratime, noatime, dir_index and > data=writeback option (work well with reiserfs) > > Benchmark: > > Hardware: Dual P IV @2.2 Ghz, 3 Gb of Ram, 2 x 10000 rpm scsi Disk (RAID1) > 900 hosts, 18700 interfaces (mostly of ethernet types so lots of 12 rrd for > each interfaces), 130 poller2 running with load average < 1. (against 70 > before). > > The database is local, I notice that it does not change anything when moving > to another server.(problems are caused by php and rrdtool). > > Cheers, > > > _______________ > David LIMA > Professional Services > www.scc.com > > -----Message d'origine----- > De : [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] De la part de Craig Small > Envoyé : mardi 20 novembre 2007 00:52 > À : jffnms-users@lists.sourceforge.net > Objet : Re: [jffnms-users] Problems after upgrade > > On Wed, Nov 14, 2007 at 04:22:56PM +0000, Martin Beecroft wrote: >> I spent the best part of 2 days trying to fix my installation of JFF, >> and now have a monitoring solution that works again. The problem was >> related to rrdtool thrashing the disk to within an inch of it's life, a >> vmstat 1 10 was showing a bi value of consistently 20,000. I fixed the >> problem by tweaking the readahead value on the disk with a "blockdev >> --setra 64 /dev/sdb1" >> >> I'm not sure if anyone else has seen similar problems, or whether this >> solution would help them, but I thought I'd best throw this out there in >> case anyone else is struggling. > > Now this does sound a little familiar to someone elses problem. I think > it was David; if it was can you confirm I'm sort of in the right area? > > I wonder if rrdtool is misbehaving? > > - Craig ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2005. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ jffnms-users mailing list jffnms-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/jffnms-users
