| I'm going out on a similar call tomorrow evening. My customer is
getting similar errors on both 2100RM's in a DSSI cluster. The
cluster communicator in this case is the EISA to DSSI adaptor,
a KFESA on one system and a KFESB on the other. I'm wondering if
this problem was ever solved. Larry Gamble is helping me dig up
data on this problem. Tomorrow evening's scheduled shutdown will be
an info- and rev-gathering session; although I've ordered a couple of
KFESB's to swap in, I don't expect much from changing hardware.
The systems run Open VMS v6.2 (no sub-rev, just 6.2). Does anyone out
there know (or even suspect) that a change to KFPSA's will fix this?
Note 1208 would seem to indicate that other DSSI adapters are
implicated. SHAC chip problem? I hope not...
Watch this space- facts to follow.
Chris Reed MCS East Providence, RI
|
| Subj: IPMT: Case HPAQ2052A (CFS.48698,Sev: 3,UMASS OF DARTMOUTH) has a new solution proposed by Mary Ann Sentance; Product: KFESA-AA
From: RISSAL::LEVANGIE 13-FEB-1997 14:49:41.87
To: PROXY::AVSRPE
CC: LEVANGIE
Subj: IPMT 48698 - Umass
Briefly, INVALID BUFFER NAME errors are a known issue that occur as a result
of an incompatability (or "race condition" as mentioned in the original case
text) in certain hardware where there is a different perception of when a
DSSI packet has been "received" as good. If a reset occurs during this time
(before BUS FREE), the initiator will resend the packet to the target that
has allready accepted the packet as "good". If the resent packet is the last
in a buffered data transfer, the retransmit attempt may not occur until after
the the target has invalidated it's buffer descriptor resulting in the error.
This appears to happen in systems at times of heavy I/O activity.
SYS$PADRIVER has been modified (as mentioned in the case) to make these errors
less "severe" by closing the VC to the node which caused the error as opposed
to closing the port. According to the maintainer of the driver, the modified
version is included in the release of Alpha AXP OpenVMS V6.2-1H2. Obtaining
the updated driver requires upgrading to that version (or above) of OpenVMS.
There is no engineering plans to make any hardware modifications for this
issue. In past cases, the updated SYS$PADRIVER has been proposed as the
solution for the condition described above. Keep in mind that the updated
SYS$PADRIVER does not eliminate the errors, it only makes them less severe.
The stars article you had mentioned in the case was a QAR and the answer was
from the maintainer of the SYS$PADRIVER.
The proposed engineering solution is to upgrade OpenVMS to V6.2-1H2 which
includes the updated SYS$PADRIVER that reduces the severity of the Invalid
Buffer Name errors.
Dave
|