From mboxrd@z Thu Jan 1 00:00:00 1970 From: erik quanstrom Date: Fri, 25 Jun 2010 23:24:11 -0400 To: 9fans@9fans.net Message-ID: <104a5f0acfa2903a5f3e04503dcffcd2@kw.quanstro.net> MIME-Version: 1.0 Content-Type: text/plain; charset="US-ASCII" Content-Transfer-Encoding: 7bit Subject: [9fans] more fun with many maches Topicbox-Message-UUID: 370c8198-ead6-11e9-9d60-3106f5b1d025 i added some tracing to tell me who's allocating what when and on what mach. a little mistake in the summary was very interesting. instead of printing the number of fast frees per mach, the total number of frees per procssor was printed. the imbalance is surprising: 128 [20/100/10] block size share 0/100 nslow 103444 nsfree 10350 nmalloc 103434 nfree 103400 0 0/0 nfast 10972 nffree 10982 1 1/1 nfast 11676 nffree 11747 2 0/0 nfast 11695 nffree 11725 3 2/2 nfast 11427 nffree 11429 >> 4 28/28 nfast 22778 nffree 126046 5 0/0 nfast 5134 nffree 5114 6 0/0 nfast 11743 nffree 11773 7 0/0 nfast 11945 nffree 11985 tot 31 nfast 97370 nffree 200801 seems almost all the allocation takes place on mach 4. this is consistent across reboot. why? - erik