Ulrich Windl wrote:
> for an rx6600 Itanium2 server which features 66, 133,1and 266 Mhz
> PCI-X slots the default configuration ships with the SAS RAID
> controller in a 66 MHz slot, as well as the Gb LAN card.

Thost would be slots 1 and 2 yes? Just to be paranoid, you do know
that the initial version of the QuickSpecs for the rx6600 (and rx3600)
were incorrect in their identification of the I/O slot specs yes? It
should be:

Slot Type
3 Independent 266 MHz
4 Independent 266 MHz
5 Independent 133 MHz
6 Independent 133 MHz
7-8 Shared 66 MHz
9-10 Shared 66 MHz

> Assuming 64bit cards, some simple math suggests that the theoretical
> peak burst transfer rate for 66 MHz is 528 MB/s. In practice it's
> maybe 1/3rd or 1/4th of it.

If it is PCI-X, with split-transaction support, I'd think that 1/3 or
1/4 is probably pessimistic. I'd probably guess 2/3d's or more -
maybe even 4/5th's although I cannot say I have hard data to back that
up, just recollections of past discussions elsewhere.

> On an AMD Opteron server with two SAS disks a RAID1 can read at 120
> MB/s roughly, and I'd expect more disks to provide even faster rates
> (unless the controller is slow).

While I have access to an rx6600 I am using for 10G Ethernet testing,
it is not setup to do iozone testing - not enough internal mechs. I'd
suggest just going ahead and running some tests to see what happens.
FWIW, netperf4 has, of all things, disc I/O tests in it now too, which
means you could do some simultaneous disc and networking tests. Of
course FTP could cause that to happen too

BTW, I just recalled that based on the flawed QuickSpecs I'd put an
AD144A 10G NIC in a 66 MHz slot, and with it I was able to get 2.6
Gbit/s on netperf tests (G == 10^9). That is ~310 MB/s (M == 1024^2)
which is just shy of 60% of the theoretical peak burst you calculated
above. I guess your stuff above is powers-of-ten, in which case that
is 325 MB/s (M == 1000^2) which is just shy of 62% - still not quite
the 2/3 I was saying above but rather better than 1/3rd .
Generally, networking has a higher ratio of control structure access
to data DMA, and the DMA's tend to be smaller than mass storage.

> As the 4Gb FC-Adapter is located in a 266 MHz slot, I wonder whether
> the RAID controller shouln't be put into a faster slot, while the
> FC-HBA might work optimal even in a 133 MHz slot.

> Any insights?

The SAS RAID and the Gb are "core" I/O.
Slots 1 and 2 are the "core" I/O slots.
ergo
The SAS RAID and Gb go into slots 1 and 2.

I doubt it was any more than that. Well, that and cable lengths and
available paths from the SAS card to the internal disc slots. Now,
whether or not the os/firmware/efi particularly care I doubt it...

What is the specific product/part number of the 4Gb FC HBA? It is
possible that it is a PCI-2.0 266MHz card and so was stuck in slot 3
or slot 4 on that basis alone.

rick jones
--
portable adj, code that compiles under more than one compiler
these opinions are mine, all mine; HP might not want them anyway...
feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...