From: Balle, S. <sus...@hp...> - 2004-08-24 18:18:54
|
John, Could you get me some information about the tests you guys did that=20 proved it to be useless compared to careful coding of the fast path? Do you have a white paper or a memo that describe the results? Thanks, Susanne -----Original Message----- From: John Levon [mailto:mo...@co...] On Behalf Of John Levon Sent: Tuesday, August 24, 2004 12:09 PM To: Balle, Susanne Cc: opr...@li... Subject: Re: Oprofile and Compaq DCPI Comparison On Tue, Aug 24, 2004 at 08:21:56AM -0400, Balle, Susanne wrote: > By compression I mean that DCPI is aggregating data/reducing the=20 > amount of data using the techniques described in the DCPI paper you=20 > have a link to on the oprofile site. No, we don't. There is no hash-table in kernel - tests proved it to be useless compared to careful coding of the fast path. There is no aggregation of samples; however, semi-persistent data (such as the current task) is only recorded once in the sample data stream. Each on-disk sample file is stored as a persistent hash table. See the internals doc on the website for more info regards john |