I am currently using collectd to collect statistic on about 100 hosts, for it I wrote my owner grapher. So far we have only run i686 and amd in 32bit and things have not been a problem. Collectd allows to run over the network but so far everytime I looked at it, I lost too many packets and so I got holes in the data. Because of that we run collectd local on each box and copy every 30 minutes the rrd files to the grapher box. This works good enough for us as long we ran only i686. But we also got 3 Solaris sparc machines now and x86_64 is our new target for Database servers.
And here lies the problem. The format of the rrd file is depend on the platform is was written on, you can only xml dump it there and then reimport on the grapher box, not very elegant nor efficent, esp. when you are start getting to larger number of machines. I really would like to see if there couldn't be support for (as option maybe) of allowing platform independ files. I know of a friend at Yahoo, who uses rrdtool also and has run into the same problem much more so then I have.
> And here lies the problem. The format of the rrd file is depend on the
> platform is was written on, you can only xml dump it there and then
Moving to a consistent structure would:
Make writing binary tools easier - skipping the expensive
Make displaying data collected by multiple platforms
and accessed via (nfs, rsync, whatever) easier and allow
a unified front end for end users. This point would be
applicable both in large scale operations, and in operations
where you're forced to work with multiple platforms.
(Or you are trying to migrate from one to another on a
medium number of machines).