Speaking of stripe sizes

Seth Mos knuffie at xs4all.nl
Tue Dec 10 12:48:01 CST 2002


At 13:31 9-12-2002 -0500, CBeck at coradiant.com wrote:
>Quick survey:
>What stripe sizes do people here use for databases with a mix of large
>(1GB+), medium (1MB+) and small (10k+) tables that write to the file system

2GB (max) extents per database and about 20GB+ all together.

>and not a raw partition?
>Which file system (Reiser vs Ext3 vs JFS vs XFS with/without LVM blah blah
>blah)

Plain XFS on a scsi device with a larger journal for more metadata updates.

>Which control system?  (Software RAID vs onboard vs attached vs SAN vs &c.)

Hardware raid 10 across 6 disks (10K rpm) using a PERC3/DC with 128MB 
cache. Chunk size is 128KB.

>Which distro?

Red Hat 7.1 with XFS bits (SGI XFS 1.0.1 installer) Running a XFS 1.2Pre3 
kernel now. (~=2.4.18-17.7.x)

>Which DB?

Progres 9.1C (9.1D in testing)

>Credit Card numbers?  ATM PINs?  (Hey, can't hurt right?)

Uhm, sorry. No can do.

>Need to make some purchasing decisions in the near future and I want to see
>what people are happy with.  I expect it's all over the map but something
>unexpected might turn up.

Creating a new database (about 3.5GB) is a excellent way to kill 
interactivity on the machine. The raid througput is about the 30-40MB/sec 
mark. A standalone disk on a scsi controller can do this as well. It just 
doesn't seem to get everything out of the hardware possible.

I am looking forward to the Perc4/DC which should arive sometime next year 
(Matt? ;-)

>PPS ... anyone with beta experience on the blade servers is very welcome to
>chime in with their experience as well.

I think they suck, but they look cool :-)
They have their uses although you can't plug in random hardware or a lot of 
disks or ...

Cheers
--
Seth
It might just be your lucky day, if you only knew.




More information about the Linux-PowerEdge mailing list