From: Johannes F. <jfo...@mu...> - 2002-11-13 21:13:55
|
At 15:52 Uhr -0500 13.11.2002, James Neal wrote: >On Fri, 2002-11-08 at 22:10, David Coulson wrote: >> Ian C. Blenke wrote: >> > Using COWs with a single backing store only seems to impede >>performance, not >> > improve it. If you're looking for performance, strongly reconsider unique >> > images for each UML instance. >> >> Interesting - Ever tried to boot 25 UMLs at once using individual files >> and comparing it to a single backing store and COWs? > >..This seemed worth an experiment. > >I used sar to watch CPU time, and recorded the time it took to go from >100% idle to 100% idle. All systems were started at the same time. > >Boot time for 22 UML images, without COW files: 9m6s >Boot time for 22 UML images, with COW files and one backing store: 3m7s Could the good disk cache using COW play a role in this test? The Data is quite identical. So there might be other results when the caching has no such a huge effect. Greetings Johannes |