On Wed, Apr 23, 2008 at 7:01 PM, Ben Rockwood <[EMAIL PROTECTED]> wrote: > The question I have for the list is... has anyone else run into this > bottleneck? If so, how did you solve the issue? I know several people > have multiple gmetad's out there, perhaps unnecessarily. CPU usage and > disk consumption are low, its just doing a lot of IO. My data/ dir is > only 27MB.
While this doesn't relate to Ganglia specifically, David Plonka, et al, wrote a paper about getting MRTG to scale; the limiting factor was disk IO. The paper is pretty interesting. Here's a link to it, and the slides for a talk based on the paper. http://www.usenix.org/events/lisa07/tech/full_papers/plonka/plonka.pdf http://net.doit.wisc.edu/~plonka/lisa/lisa2007/bigmrtg-lisa-talk.pdf Also, this email post/thread is interesting as well: http://www.mail-archive.com/[EMAIL PROTECTED]/msg01887.html >From the article, the three "easy things" you can do are: 1) use rrdtool 1.2.24 or later (it has the fadvise patches) 2) use less aggressive read-ahead 3) use a separate partition for the rrd files, with various mount options (noatime, dir_index, etc) Many people use tmpfs, and that certainly seems to be a common and fairly simple solution. -- Jesse Becker GPG Fingerprint -- BD00 7AA4 4483 AFCC 82D0 2720 0083 0931 9A2B 06A2 ------------------------------------------------------------------------- This SF.net email is sponsored by the 2008 JavaOne(SM) Conference Don't miss this year's exciting event. There's still time to save $100. Use priority code J8TL2D2. http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone _______________________________________________ Ganglia-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/ganglia-general
