[Linux-cluster] Low cost storage for clusters

[Riaan van Niekerk]

Patton, Matthew F, CTR, OSD-PA&E wrote:
  >  > I'm happy enough that iSCSI is acceptable, I don't think I
>  > will be able
>  > to justify a full fibre-channel SAN. Certainly I'd expect it to do as
>  > well as straight SCSI320 given that it has 3x the raw bandwidth (even
>  > accounting for the TCPIP overhead).
> 
> use Jumbo frames and thruput might be much better. Then again I don't 
> know how iSCSI generates packets. I hope it's not 1 packet per disk 
> block which is typically 512bytes. presumably how many disk blocks / 
> packet is a (auto-)tunable parameter.

I have often heard that jumbo frames is a very easy "low hanging fruit" 
for performance optimizing of iSCSI. I do not have any experience with 
it myself tho.

> switch price matters. But some of the cheepies claim features that the 
> big-names charge 10x as much for which doesn't make sense. I never pay 
> the Cisco tax unless there is a gun to my head.

Puts a new meaning to "Nobody ever got fired (at) for buying 
IBM/Cisco/Oracle"

my 2c regarding hardware initiators (HBAs) vs SW initiator - this was 
mentioned in this or another thread today:

I have also heard anecdotal evidence that hardware initiators are not 
all that much faster than software initiators.

1 The reasoning behind it being similar to why software RAID 1 (with a 
big, idle Xeon behind it beats the pants of dedicated hardware RAID 1 
with its relatively underpowered ASIC.

2 Also, the TCP offload engine (TOE) in a hardware initiator should give 
it an edge, but most server NIC vendors are starting to bundle TOE in 
their regular NICs, negating this performance benefit of dedicated iSCSI HW.

If possible, get a demo QLA iSCSI HBA and test the two.

Riaan
-------------- next part --------------
A non-text attachment was scrubbed...
Name: riaan.vcf
Type: text/x-vcard
Size: 310 bytes
Desc: not available
URL: <http://listman.redhat.com/archives/linux-cluster/attachments/20060816/2ee6641e/attachment.vcf>