You're right. I won't bother to try it. :-) The article didn't mention tmpfs so I assumed they are writing 300,000+ rrd files to real disk. However, they do this somewhat sequentially every 5 minutes. Ganglia, on the other hand, is writing all over the place continuously if you have a few dozen gmetads polling every few seconds. I don't like the idea of sending the head all over the hard drive constantly. So tmpfs with periodic sync to disk will still be the most efficient way to go.
-twitham -----Original Message----- From: Carlo Marcelo Arenas Belon [mailto:[EMAIL PROTECTED] Sent: Thursday, March 13, 2008 3:53 PM To: Witham, Timothy D Cc: Jesse Becker; Martin Hicks; Ganglia-general@lists.sourceforge.net Subject: Re: [Ganglia-general] Setup large clusters On Thu, Mar 13, 2008 at 01:20:31PM -0700, Witham, Timothy D wrote: > Now that is way interesting. Maybe if you do that, you can put a huge > grid on a real disk instead of tmpfs? I might have to try that... rrdtool 1.3 (currently in beta) also includes MMAP support and madvise (for file system cache optimization) that could help as well. IMHO running it on tmpfs with a 64bit OS is the most scalable way (RRD usage is tiny and memory is cheap). and syncing the files from memory/disk periodically is all that is needed for reliability. Carlo ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Ganglia-general mailing list Ganglia-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/ganglia-general
