[SERVER-663] need to test memory mapped file performance on freebsd Created: 23/Feb/10  Updated: 06/Dec/22  Resolved: 30/Dec/15

Status: Closed
Project: Core Server
Component/s: Storage
Affects Version/s: None
Fix Version/s: None

Type: Bug Priority: Major - P3
Reporter: Dwight Merriman Assignee: Backlog - Storage Execution Team
Resolution: Won't Fix Votes: 7
Labels: freebsd, fsync, mmap, msync
Remaining Estimate: Not Specified
Time Spent: Not Specified
Original Estimate: Not Specified

Issue Links:
Related
Assigned Teams:
Storage Execution
Participants:

 Description   

see http://groups.google.com/group/mongodb-user/browse_thread/thread/e0a5b279a6389d2c/186f153e39ee2fe5?hl=en&lnk=gst&q=freebsd#186f153e39ee2fe5



 Comments   
Comment by Reed O'Brien [ 06/Sep/12 ]

Somewhere in one of those email threads, specifically this one:
http://lists.freebsd.org/pipermail/freebsd-stable/2011-December/065108.html

It appears a patch was made to help – hopefully fix – this. It looks like it missed the 9.0 release but is landing in 9.1:
http://svnweb.freebsd.org/base/releng/9.1/sys/vm/vm_object.c?r1=225736&r2=229695

I haven't tested it yet, but have my fingers crossed.

Comment by David Dodd [ 13/May/12 ]

We are operating a production master/slave pair running FreeBSD 8.2 running mongo 2.0.2 and the master server is infrequently becoming unresponsive (once every two to three weeks).

db.currentOp() has been reporting the slave replication connection as the active process during the problem.

We have seen similar symptoms during the problem.

Has anyone verified the efficacy of the patch with regard to journalling ?

Comment by Scott Sandler [ 20/Apr/12 ]

I have done extensive load testing on freebsd 8.2, mongo 2.0.2, with and without the patch mentioned in Scott's post. I have found that I can very consistently get to about 10x the concurrency with the patch vs. without. In either case global lock queueing is the first bottleneck. These tests were done with journaling off (we originally had journaling on, but found turning it off helped mitigate a bit of the lock queueing we were seeing).

Comment by Alvin Richards (Inactive) [ 30/Jan/12 ]

And discussed here

http://www.mail-archive.com/freebsd-stable@freebsd.org/msg118225.html

Comment by Scott Hernandez (Inactive) [ 16/Dec/11 ]

It seems from this discussion that using fsync is more appropriate on freebsd. Not sure about journaling ramifications.

Patch: http://zonov.org/mongodb/patch-util-mmap_posix.cpp
Discussion: http://lists.freebsd.org/pipermail/freebsd-stable/2011-December/064844.html

Comment by Johnny Boy [ 23/Jun/11 ]

FreeBSD 8.2 with mongo 1.8.1 doesn't seem too quick with multiple inserts either, actually 10+ times slower than other machines running mac and linux

Comment by Philip Southam [ 11/Jan/11 ]

The behavior outlined in the referenced discussion thread still seems to be a problem on 1.6.5 running within FreeBSD 8.1.

Generated at Thu Feb 08 02:54:48 UTC 2024 using Jira 9.7.1#970001-sha1:2222b88b221c4928ef0de3161136cc90c8356a66.