[lkml]   [2012]   [Jul]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH 00/40] AutoNUMA19
On 06/28/2012 08:55 AM, Andrea Arcangeli wrote:

> o reduced the page_autonuma memory overhead to from 32 to 12 bytes per page

This bit still worries me a little.

Between the compact list management code, the migrate
daemon, the list locking and page queuing and unqueuing
code, we are looking at 1000-2000 lines of code.

I know your reasoning why asynchronous migration done
by a kernel thread should be better than migrate-on-fault,
but there is no actual evidence that it is in fact true.

Would it be an idea to simplify the code for now, and
add the asynchronous migration later on, if there are
compelling benchmark results that show it to be useful?

The amount of source code involved is large enough that
a justification like that would be useful, IMHO...

All rights reversed

 \ /
  Last update: 2012-07-09 18:21    [W:0.502 / U:1.108 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site