Lustre devel - Apr 2007 - [Bug 12329] New: MGS/MDT on single node cant handle lots of OST's mounting simultaneously

Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=12329



During some many ost tests i ran this month i ran into a problem where i cannnot
mount formatted OSTs very quickly.  Here is my setup:

All nodes are dual IA64 1.5Ghz nodes.
570 OST''s with a RAID0 OST each. 

I format and mount the mgs/mdt as such on m978:
mkfs.lustre --reformat --mgs --comment "TestMGS" /dev/sda7
mkfs.lustre --reformat --mdt --fsname=mpp2_r0 --comment "Really Big OST
Test MDT
raid0" --mgsnode=978@elan0 /dev/sdb7

mkdir /l
mkdir /l/mgs
mkdir /l/mdt

mount -t lustre /dev/sda7 /l/mgs
mount -t lustre /dev/sdb7 /l/mdt



I format the ost''s as such:
$MKFS --ost --index=$num --fsname=$fsname --mgsnode=978@elan0 --comment
"$(hostname) md9" --reformat /dev/md9

where $num is based on the node number(0-569)

then I attempt to mount the ost''s with a pdsh:
pdsh -w m[0-569]  mount -t lustre /dev/md9 /l/md9

almost immediatly the MGS and OST nodes start complaining.  This happens in the
first 32 nodes(pdsh default fan-out)  ( i''ll attach console logs to the
bug
momentarilly)

If i reduce the pdsh fanout to 8 things work fine, but it takes 10-20 minutes to
mount lustre ost''s

I saw this problem before in 1.5.95 wit 7 OST''s per OSS, and had to
reduce
fanout to 2.