Slides are here: http://www.kitchensoap.com/2007/04/27/slides-from-capacity-planning-for-lamp-talk-at-mysql-conf-2007/
A lot of folks at the talk asked about rrd storage, and some folks nodded when I mentioned we store our rrds on ramdisk and then sync them to disk periodically. Also a couple of people mentioned hacking on the templates, so I thought I'd mention what we've done at Flickr: - put a stacked graph of the chosen metric *above* the individual node rrds in a cluster. This way, we can see things like "total hits" without having to wait for >100 nodes to draw their own graphs. - changed the web frontend so you can put arbitrary time values, like h,m,d,w,m for hours, months, etc. So: "r=8h" will show you 8 hours. - added the custom graph add-on (thank you, Alex you deserve a beer bought for you) (from here: http://wtf.ath.cx/screenshots.html) My impression is that a lot of folks in the web world use ganglia, and have written again and again some of these changes to make it more useful for them. I'm almost sure that the majority of hacks are on the interface, not really with gmond or gmetad code. We have custom gmetrics for measuring Squid use, memcached, MySQL, Apache, even OS level metrics (context switches, for example) that others would find super useful. I realized at the talk that I got the first version of my memcached gmetric script from someone at Facebook, secondhand, before I added to it. Is there any plans to revive the gmetric repository ? I think a lot on this list would contribute to it. I would. -john __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com