[Gluster-devel] Out of memory problem with 1.4.0qa5

Lukas Hejtmanek xhejtman at ics.muni.cz
Tue Oct 14 08:43:58 UTC 2008


Hello,

I'm using NUFA scheduler with read-ahead, write-behind and io-cache
translators. I'm using x86_64 architecture. I have 3 servers that act both as
clients and servers.

I run bonnie++ benchmark. The machines are equipped with 4GB RAM so bonnie by
default use 8GB test file split into 8 1GB files. When doing read performance
tests, io-cache translator mmaps *all of eight 1GB files* into the RAM which
causes system outage as there is not enough free RAM or swap. This nasty bug
is present only on 64bit architecture because IA32 does not allow mmaping more
than 2GB space. 

-- 
Lukáš Hejtmánek





More information about the Gluster-devel mailing list