[Bioclusters] Random bits: GigE copper eval; biocluster pictures & IBM storage problem

[Chris Dagdigian]

Hi folks,

Several unrelated bits today...

(1) Slashdot linked to a recent eval of gigabit-over-copper HBA cards 
which was great reading for me since it looks as though I'm going to 
have to build  a box capable of doing NAT and intrusion detection 
between at least 2 and possibly more copper gigabit network links.

The slashdot story is at 
http://slashdot.org/articles/02/04/14/2124257.shtml?tid=126

The actual review is at http://www.cs.uni.edu/~gray/gig-over-copper/

Any comments? experiences with fiber vs copper for GigE? My current 
project is going to use lots of fiber and copper gig connections going 
into an ExtremeNetworks Alpine 3808 switch so I may be in a position to 
try some experiments over the next month or two.

(2) Harvard gave me permission to publicize the pictures I've been 
taking during the hardware build process at the new Bauer Center for 
Genomics Research (http://cgr.harvard.edu) . Basically they are just 
getting started and I've been involved in helping sort out the initial 
research computing infrastructure which basically boils down to: 4TB 
NetApp NAS + 60 CPU Linux cluster running Platform LSF + 360-tape 
SAN-attached AIT tape library robot and a bunch of misc. support 
systems. The datacenter is still under construction so I've been 
building this stuff in an office over the last two weeks.

Pictures from this effort along with pictures from the Vertex Pharma 
VAMPIRE cluster and Steven Brenner's system at Berkeley are all online 
at http://gw.sonsorol.net:8080/gallery/bioclusters

That site may not be super reliable as gw.sonsorol.net is hanging off a 
cable modem instead of a true dedicated link.


(3) IBM storage problem

http://www.storage.ibm.com/hardsoft/products/fast200/fast200.htm

I'm currently trying to figure out why a fibre-channel FastT200 storage 
server from IBM is performing slower than the internal SCSI-based 
ServerRAID disks.  I'm looking for pointers that would help me sort this 
issue out as well as any and all info on what sorts of real world IO 
performance I should actually get out of a FastT200 system with a single 
raid controller and single fibre-channel connection.

The server in question is an IBM x340 server. It has 3 internal SCSI 
disks at RAID5 and has a Qlogic 2200 FC HBA that allows it to be 
connected straight into the FastT200 storage controller. The FC array 
has a single shelf with 10 drives which have been split into two RAID5 
volumes. There is nothing redundant about it: single controller, single 
connection to the host and no switch or SAN stuff in the middle.

The central problem is that running iozone and bonnie on both the SCSI 
and FC volumes shows that the SCSI arrays is significantly faster than 
the volumes that are mounted via the fibre-channel connection.

Anyone seen anything similar?