I/O bandwidth is the key. Since these new systems cannot match the I/O
bandwidth of the AIX pSeries systems they are replacing, count on having
to run MORE SERVERS to accomplish the same work.

Having said that, there's a silver lining. Most TSM systems could
benefit from subdivision. The database becomes more manageable.
Expiration Angst is reduced. You can run the Log in ROLLFORWARD mode
again. AUDIT DB once again becomes feasible. Opportunities open up for
having TSM servers back each other up. Now is a good time to do this,
ESPECIALLY if you are going to make the move with EXPORT TO SERVER.

A fallback plan, in case the new servers' performance doesn't measure up
to expectations, would be to initially plan each new server to have two
TSM server images and two databases running on it. Then when everything
bogs down, you can call Dell and bring in more machines, without having
to go through EXPORT all over again.

So, plan now on having more servers and more server images, partially to
compensate for the new systems' poorer I/O bandwidth, and take it as an
opportunity to do something that you probably should have considered for
a while. If the new systems simply don't work and you convince them to
go back to AIX pSeries, you'll still be ahead having divided your server
images.

Roger Deschner      University of Illinois at Chicago     rogerd AT uic DOT edu


On Tue, 9 Jan 2007, Bob Booth wrote:

>On Tue, Jan 09, 2007 at 09:59:20AM -0500, Zoltan Forray/AC/VCU wrote:
>> Windows is not in the picture. These are going to be RH Linux systems.
>>
>> We will probably perform EXPORT TO SERVER to move the clients, one by one,
>> since we are probably going to move to new media types (3592 vs 3590) , as
>> well.
>
>Look for the type of system that will give you the most PCI-X bandwidth you
>can get.  You are going to be pumping quite a bit of data across that bus,
>with presumably 4Gb HBA's, some SCSI (?), networking...  TSM is all about
>I/O, to disk, tape, and the network.
>
>hth,
>
>bob
>