On Sun, 2009-11-15 at 21:25 +0000, Bishop, Peter wrote: > I'm still a bit leery of the extra I/O, as that equates to elapsed > time which is our enemy in this scenario. I may try a named pipe test > on a small case just to see what happens but Leslie's small test does > not augur well.
Look at Leslies follow up post - I suspect the initial results were invalid. I don't understand why you think there will be "extra" I/O. Any reads will go through the page cache. The data are referenced there - by both (any) programs. The named pipe will stall waiting for the reader process. Unless the cache is flushed and the data subsequently re-read, there will be no more than 1 I/O. If the files are small and recently created, they may even still be *in* the cache. Unlikely if you follow the (zVM) recommendation to screw the guest memory allocation down as much as possible. But again, the guest may see them as real physical I/O (FSVO "real" in this context), and zVM may still have the pages in storage. So no (disk) I/O. If the application is doing (synchronous) raw/direct I/O, ignore all the above. All bets are off in that case. Shane ... ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390