Hi,
I must confess, I am not subscribed to this mailing list. I read it
via the HTML web interface and the web interface strips out the email
addresses. When I want to respond to a post, I ask my Working Leader to
forward the post to me, but he's not around today. I guess I should
just create a separate account for the Tru64-UNIX-managers list, and
then subscribe to the list with that account. I'll get around to it.
Anyway a while back Halllstein Lohre posted to the list "Slowdown after
V5.1/TruCluster patch kit 3" but I never got around to asking my W/L
for his email address and now he is on vacation today.
Anyway Hallstein Lohre writes:
"The backup was running at about 6 MB/s, but now we are seeing
close to 1 MB/s. This leads to problems with the backup
window, and at the moment backup is the biggest problem"
"I feel a bit stuck here..."
I looked for his summary but I didn't see one, and I'm not sure if he
resolved his problem.
I applied a patch kit-003 to a TRU64 5.1 Alpha1 client today. Before
I applied the patch I ran a Networker backup to benchmark my speed.
The backup server runs TRU64 UNIX 5.1 with no patch kits installed and
Legato Power Edition release 6.02 build 251. I used the Networker GUI
to set parallelism for the "test" client to (8), and target sessions are
set to (4) for the devices so my test only used two of the TL895's seven
drives which makes it kinda look like a TL891 with two drives. :) Most
people identify with the TL891. Lohre is clustered so he should be
running with Legato's Power Edition too. Some of my Windows NT servers
are clustered and that is why I am running with the Power Edition
instead of Network Edition. At least some of my configuration matches
Lohre's. :)
Speed before my patch kit application hovered around
/dev/ntape/tape5_d1 4097 KB/S 4 sessions
/dev/ntape/tape6_d1 4097 KB/S 4 sessions
After the patch kit was applied 5.1 patch kit 003 I ran the backup
again
Speed after my patch kit application hovered around
/dev/ntape/tape5_d1 4161 KB/S 4 sessions
/dev/ntape/tape6_d1 4161 KB/S 4 sessions
That's about as fast as my environment supports today. Here is my
topology: The "test" client's data is stored on an EMC Symmetrix data
array where it travels into a KZPBA-CB UltraSCSI storage adapter onto
the client's box, then out the client's 100baseT NIC card, around a
smart switch, and then back through the server's 100baseT NIC card where
it gets distributed amongst four KZPBA-CB ULTRASCSI storage adapters
modules driving the TL895.
(* Next fall we will be upgrading our important servers for fiber and
gigabit ethernet *)
I only applied patch kit 003 to my "test" client, not to the production
backup server, also I am not TruClustered and this may be why I am not
showing any slow downs in speed.
However here is my theory on Hallstein Lohre's problem. I'll pretend
his backups are going out through the client's ethernet interface around
the smart switch and then back into the server's ethernet interface.
However it could be one of those new fiber only deals, ATM deals, or
just a straight backup directly from disks attached to the server in
which case my theory is useless but I am proposing it is an "ethernet
problem" not a "patch kit" problem.
Here it is. Once I clobbered performance much like Hallstein Lohre
reports playing around with the duplex settings on my ethernet cards. I
have tried settings: half duplex, full duplex, and auto negotiate on
two different types of ethernet cards. My older NIC cards do not
support the auto negotiate feature. They work best when 100baseT
half-duplex is forced. When I try forcing full-duplex on those NIC
cards it clobbers performance. Performance drops from 100baseT to
10baseT. I made a few inquiries as to why this might be happening and I
was told when forcing settings on ethernet cards quite often you must
also force a simular settings at the smart switch too. Eeeek, I don't
know how to do that, so I quit using full-duplex on my older ethernet
cards and my performance problems went away. It was amazing. Now
then, my newer NIC cards do support the auto-negotiate feature and they
do auto negotiate for 100baseT full duplex operation successfully with
the smart switch and no performance problems are noted with the newer
devices. The auto negeotiate feature must be user selected though.
Hope this helps, sorry I couldn't send it via private email. Oh, and
Eeeek, I'm not TruClustered, sorry for butting in with my non clustered
results.
Good Luck
Kevin Criss
Received on Fri Jun 29 2001 - 23:00:40 NZST