Hello, > >I have a pair of Linux machines graphing a lot of switch ports > >(~35000 ports across 1500 devices each) on 5-minute intervals. > >Disk I/O seems to be my biggest problem. The CPU load tends to
I have few additional commentaries: > We have a similar problem. I find these help: > 1) FAST DISKS. Get the fastest disk IO you can. Nice quick SCSIs or SAN. > IDE just doesn't cut it. I use two 15000RPM Ultra320 SCSI disks in a mirror. I use MegaRAID SCSI controller. > 2) Mount the filesystem with noatime option. This speeds it up a lot as > the atime flag doesn't have to be updated. It's good tip. > 3) Splt RRDs into multiple filesystems. A big directory is a timewaster. > This is less significant, though. You can try tune filesystem. I enabled "dir_index" on ext3(2) filesystem which resolve problems with big directories. This can be done on live filesystem with tune2fs. dumpe2fs /dev/sda4 | less Filesystem features: has_journal dir_index filetype needs_recovery sparse_super > 4) Lots of memory. UNIX will use all spare memory for disk cache. I have 3GB of RAM. All files are cached in memory during measuring. I have about 8000 targets. MRTG runs for about 30 seconds, when all targets are available. Flushing cache to disk takes about 15 seconds. I use Linux with kernel 2.6.9. momentally. Best regards, Pavel Ruzicka > 5) Use RRDTool, not log files! Of course, you are doing this already. > 6) Tune the filesystem for smaller files and small block IO. You never need > more than the size of an RRD to be loaded, but you want it to block-load > the whole RRD file. -- Unsubscribe mailto:[EMAIL PROTECTED] Archive http://www.ee.ethz.ch/~slist/mrtg FAQ http://faq.mrtg.org Homepage http://www.mrtg.org WebAdmin http://www.ee.ethz.ch/~slist/lsg2.cgi
