Karl Hancock ([log in to unmask]) wrote:
: Are there any options to put a 100BaseT card in an HP995 and a
: HP9000 G40?

For the 9K G40, there is an HP-PB 100BT card, the glossies should be
somewhere on http://www.hp.com/. I'm not sure which HP-UX revisions
are supported.

I wouldn't expect 100 Mbit/s from it though. While 100BaseT ups the
raw bandwidth on the wire, it does *nothing* to improve host
efficiency of sending data (it keeps the same tiny MTU of Ethernet).
So, if you have a system that takes 25% of the CPU to send at 10
Mbit/s, you will not get above 40 Mbit.

I've not benchmarked an HP-PB 100BT card in a G40, but I would be
surprised to see more than 40 Mbit/s from it. 30 might be more like
it.

Contrast that with HP-PB FDDI - FDDI ups the MTU by 3X or so, which
makes sending bulk data *much* more efficient. There are some FDDI
numbers for various system (not G40's though) in the netperf database
(http://www.cup.hp.com/netperf/NetperfPage.html).

Past experience has indicated that if you drop the MTU of FDDI to 1500
bytes, it will increase the CPU util of something like a TCP_STREAM
test by 50%. You can use that when working-out a fudge factor for
estimating 100BT performance. The HP-PB FDDI card also provides
checksum offload, which is probably another 10% delta. Some of the
features of that card (under HP-UX at least, I don't know if MPE uses
them all) can be found at:

   ftp://www.cup.hp.com/dist/networking/briefs/copyavoid.ps.Z.

That document has some data for G30 class CPU, and indicates that a
G30 running HP-UX 9.0 could get ~57 Mbit/s using virtually all of the
CPU. Take away the checksum offload, and it drops to 53 Mbit/s. Drop
the MTU to 1500 bytes and increase the CPU util by 50% and you arrive
at about 35 Mbit/s.

Regardless of system I don't think anyone has clocked the HP-PB 100BT
card over 50 Mbit/s. If someone manages it, the netperf database would
love to see the result :)

rick jones
http://www.cup.hp.com/netperf/NetperfPage.html