| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 1233.1 | | STAR::GAGNE | David Gagne - VMS/LAN Development | Thu Feb 10 1994 14:33 | 5 |
| Try using some tools to determine if it's the network or the system
that's the bottleneck.
Check your LRPSIZE/LRPCOUNT SYSGEN parameters on all the nodes - if
you're running a system where these count.
|
| 1233.2 | Here's What They Have | KZOVAX::STOUGHTON | | Fri Feb 11 1994 12:37 | 15 |
| Currently all three nodes in question have a lrpsize of 1504 and a
lrpcount of 1024.
The interesting thing is that when the systems were running debnas
there were no problems since the addition of the FDDI, Only the 6460
and 6410 run slow. The 6620 runs fine.
The customer is upgrading to 7.2 firmware for the FDDI, replacing
the DECconcentrator 500 module, and is updating Pathworks to the latest
rev.
The last time the concentrator was powered down the problem seemed to
clear up for a few days.
Gregg
|
| 1233.3 | | STAR::GAGNE | David Gagne - VMS/LAN Development | Fri Feb 11 1994 13:56 | 8 |
| There LRPSIZE should be in the 4000-5000 range if you want the same
performance as you had before FDDI. With such a small LRPSIZE, the
FDDI data structures aren't coming out of the look aside lists; instead
they're coming out of regular non-paged pool which can be CPU costly.
There are some articles available about setting LRPSIZE for systems
with FDDI adapters. I would assume that the CSCs have one available.
This applies only to OpenVMS VAX V5.5-2 and earlier.
|
| 1233.4 | | NPSS::WADE | Network Systems Support | Mon Feb 14 1994 09:11 | 6 |
| LRPSIZE should be => 4541; see blitz #1341, March 11, 1993.
The LRPSIZE is at least partially to blame for the perf prob.
I don't know why the 6620 isn't showing the problem.
|
| 1233.5 | THANK YOU... | KZOVAX::STOUGHTON | | Tue Feb 15 1994 11:35 | 8 |
| I found out that the 6620 IS having the problem. It is not as quickly
noticable.
We will be reseting the parameters tonight on a test cluster.
Thanks for all the help,
Gregg
|
| 1233.6 | | STRWRS::KOCH_P | It never hurts to ask... | Wed Feb 23 1994 15:41 | 8 |
| What kind of protocols does this problem relate to?
I have a customer with 4 10/100 bridges for 8 floors of a building. The
connection for all the Ethernet systems is FDDI via a VAX-6000 model
620 with a DEMFA. There are running OpenVMS VAX V5.5-2. Does this LRP
size affect everything?
Can someone post the "BLITZ" message for all to see?
|
| 1233.7 | | STRWRS::KOCH_P | It never hurts to ask... | Fri Feb 25 1994 14:05 | 1 |
| Can't anyone help out -.1?
|
| 1233.8 | | NPSS::WADE | Network Systems Support | Fri Feb 25 1994 16:13 | 78 |
|
An easy way to see if you are having the problem is to go into sda and
see if your 4896 byte VCRPs are allocated from non-paged pool
sda> show pool/typ=vcrp/head
The blitz -
From: CSSE::PATTENDEN "BARBARA PATTENDEN, TECHNICAL INFORMATION MANAGEMENT, 276-9713 15-Mar-1993 1217" 15-MAR-1993 12:18:22.89
To: @TD
CC: PATTENDEN
Subj: SYSGEN Parameter LRPSIZE must be Incremented for ALL FDDI equipped VAX Systems Running VMS version 5.4-3 through 5.5-2.
+---------------------------+TM
| | | | | | | |
| d | i | g | i | t | a | l | TIME DEPENDENT BLITZ
| | | | | | | |
+---------------------------+
BLITZ TITLE: SYSGEN Parameter LRPSIZE must be Incremented for ALL FDDI
equipped VAX Systems Running VMS version 5.4-3 through
5.5-2.
DATE: March 11, 1993
AUTHOR: Bill Wade TD #: 1341
DTN: 226-6271
ENET: VCSESU::WADE CROSS REFERENCE #'s:
DEPARTMENT: VAXcluster SASE (PRISM/TIME/CLD#'s)
INTENDED AUDIENCE: ALL PRIORITY LEVEL: 1
(U.S./EUROPE/GIA)
(1=TIME CRITICAL,
2=NON-TIME CRITICAL)
=====================================================================
PROBLEM:
There is a high probability that VAX Systems running VMS version
5.4-3 through 5.5-2 and equipped with an FDDI adapter(s) are running
with an incorrectly sized LRPSIZE parameter value. The result is
receive buffers are allocated from general pool rather than from
the LRP lookaside list which results in a degradation of system
performance.
If the LRPSIZE parameter value is not large enough for the data
structures holding the incoming messages, the buffers in the data
link driver's receive ring will be carved out of nonpaged pool
rather than LRPs. Allocation from nonpaged pool is much more
expensive than from the LRP lookaside list and the nonpaged pool
allocation is done at ipl 11 while the data link drivers do most of
their work at ipl 8.
The release notes for VMS 5.4-3 specify a value for the LRPSIZE
parameter of 4474 for VAX Systems equipped with FDDI. This value
must be increased by 131 (4605) for VMS version 5.4-3 and 67 (4541)
for VMS version 5.5 thru 5.5-2. This is ONLY required for FDDI
equipped VAX Systems running VMS version 5.4-3 through 5.5-2.
RESOLUTION:
All FDDI equipped VAX Systems running VMS version 5.4-3 should have
the LRPSIZE parameter set to a mimimum of 4605. All FDDI equipped
VAX Systems running VMS version 5.5 through 5.5-2 should have the
LRPSIZE parameter set to a mimimum of 4541. This SYSGEN parameter
change requires that the system be rebooted.
ADDITIONAL COMMENTS:
This adjustment may require an increase in LRPCOUNT as more LRPs
will be consumed. This can be determined by monitoring the LRP
lookaside list using the $ SHOW MEMORY/POOL/FULL command.
Autogen should be run with the feedback option if the current size
for the LRP lookaside list has increased beyond the initial size.
*** DIGITAL INTERNAL USE ONLY ***
DIGITAL INTERNAL USE ONLY Document
|