Timwi wrote:
Nicolas Weeger wrote:
What do you think of that ?
I don't know what hardware we'd choose, but I wholeheartdly agree on the need for some better hardware than what we have now. During afternoon (french time), it becomes slow as hell, painful to see Wikipedia crawl like that... -.-
Buying more and more hardware isn't always the best solution, especially not when "it's slow" is all you know.
Timwi
It would be useful to have MRTG or similar graphs available for _all_ the servers in the cluster, showing for each system: * disk I/O seeks/second * disk I/O bytes/second * network interfaces bytes/second * CPU load * memory usage * swap load
This, in conjunction with the existing real-time performance graphs, would enable "many eyes" to inspect the system, and form hypotheses about where the bottlenecks are.
-- Neil