| Steph 2006-03-21, 11:23 am |
| Hi group,
I wonder if anyone can help me/provide some insight into a problem I've
got with ASE 12.5.3 on linux. I've Googled all over the place and can't
find anything recent or relevant.
We've got some new hardware that we're going to be migrating our database
onto. (from old faithful Solaris SPARC boxes onto spiffy new Opterons
running RHEL 64-bit). All of the cross platform stuff seems fine, but I'm
seeing a *really* odd issue when trying to create the database devices.
If we do the disk init for all of the devices using the default memory and
engine settings, it creates the disks just fine. If we change the
parameters as we hope to use them when the DB goes live (configuring max
memory to something large, say over 4GB and with 3 engines), the disk init
grinds to a stop after it's got a fair way into the device creation.
We're creating devices on an ext3 filesystem, itself on a RAID-10 array.
When it starts the disk init, it flies, taking around 4 minutes to create
a 16GB device. When the problem is triggered (around 100 devices into the
creation) the write rate drops to around 20-30kb/s. There is no steady
degradation, it happens suddenly. I think the array is ok, cos I can
create nice big files very quickly with dd on that volume whilst this is
going on.
top says that the engine doing
the creation is pegging that CPU at 100% but suspiciously it's doing it in
kernelspace not userspace (0% 'us', 100% 'sys'). This makes me wonder
whether I have a linux memory problem. The box has 16GB RAM and is
configured with 16GB swap (after I read that there are problems with
committing large amounts of memory as we want to do for the shared
memory). I have shmmax set to 12GB at the moment, so assigning 'max
memory' of 4GB shouldn't be any issue should it?
I'm just wondering if any of you guys have ever seen anything like this
before.
Any comments/suggestions/flames for being an idiot and missing something
would be greatly appreciated
Many thanks for your time,
Steph
|