I agree. Storing metrics on Ramdisk/tmpfs is the way to go. If you think about it there are 1600 hosts with minimum 30 metrics per host which is 50k metrics that could be updated couple times a minute. Lot's of I/O load. Ben Hartshorne has posted scripts on how to automate copying from tmpfs/ramdisk to permanent storage and vice versa.
http://ben.hartshorne.net/ganglia/ You could also always point multiple gmetad processes (on different nodes) against the head gmond. That way you have some redundancy. Vladimir On Fri, 4 Dec 2009, Bernard Li wrote: > This is most likely an I/O related issue. So please try Ofer's > suggestion by putting your RRD files onto tmpfs. An alternative > solution is to put the RRD files onto a (small) RAID that could > provide higher I/O, or potentially SSDs (if cost is not an issue). > > The code from trunk does support the new rrdcached feature that comes > with newer versions of RRDTool, but that code is currently still in > development. > > Please let us know how it goes. ------------------------------------------------------------------------------ Join us December 9, 2009 for the Red Hat Virtual Experience, a free event focused on virtualization and cloud computing. Attend in-depth sessions from your desk. Your couch. Anywhere. http://p.sf.net/sfu/redhat-sfdev2dev _______________________________________________ Ganglia-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/ganglia-general
