Possible memory hw-error

[john korterman]

Hi everyone,

on a model ia64 hp server rx8620, running B.11.23, we often se Sybase crash.

Sybase suggests that defective hw, especially memory, may be the cause.

However, there are no indications in syslog/dmesg of any hw-problem.

In another thread I saw Bill Hassel recommend the below command, which I execued on the machine:
# echo "selclass qualifier memory;info;wait;infolog" | cstm

and found this in the output:

Memory Error Log Summary


DIMM Location Error Address Error Type Page Count

---------------------- ---------------- ---------- ------------- -----

Cab 0 Cell 3 DIMM 1A 0x2063a3380 Single-Bit 0x2063a3 21

Cab 0 Cell 2 DIMM 3B 0xea5e80700 Single-Bit 0xea5e80 3

Cab 0 Cell 0 DIMM Ô· 0x81883cc78000a7b8 Multi-Bit 0x81883cc78000a N/A


Would you judge the above as hw-errors in need of repairing?

regards,
John K.

[Steven E. Protter]

Shalom,

Test memory with the excercize function of cstm mstm or xstm.

Also check dmesg output.

With these tools you will find the problem.

SEP

[A. Clay Stephenson · Accepted solution]

The single-bit errors are no cause for concern because the CRC mechanisms can correct these "on the fly" with no impact to applications and the OS but the multi-bit error is more serious --- I would definitely have that DIMM replaced. As long as the single-bit errors are infrequent (~ 1 per week or so) then I wouldn't worry about them. SBE's can actually be caused by things such as background radiation so it is unrealistic to expect zero SBE's over long periods of time ---- but they should be very rare.

[tkc]

check the file /var/opt/resmon/log/event.log, how frequent is the memory events logged?

all memory events are given in :
http://docs.hp.com/en/diag/ems/memory_ia64.htm

check the PDT and see how many % usage too.

if too many SBE, system may be slow because of correcting the SBE in the memory too often.

[tkc]

oh ya, i noticed you have multibit error too? don't you have a system crashed due to this multibit error? any system panic or reboot that you didn't realise?

[tkc]

if what i have suspected your system crashed due to the memory multibit error, then i would like to suggest you to collect the mca files located in /var/tombstones directory and have the mca file dated on the day the system crashed and have it sent to hp for analysis. hp would be able to provide you the answer for the crash.

[Torsten.]

Even a multibit error may not lead to a chrash because of the chip spare/chip kill technology in sx1000 systems like yours. This is even improved for systems using the sx2000 chipset.

Read about it (manual is for different systems, but it applies more or less):

http://h71028.www7.hp.com/ERC/downloads/rx4640_2600_1600_wp_FINAL_4-14-04.pdf?jumpid=reg_R1002_USEN

page 27
"ECC and chip spare memory"

more for sx2000 systems:

http://h71028.www7.hp.com/ERC/downloads/c00767235.pdf?jumpid=reg_R1002_USEN

Anyway, a single bit error may not be a problem. Regarding the multi bit error:
Let HP support analyze your logs. They will you help further.

[john korterman]

Thank you very much for your quick responses!

As mentioned, there are no indications of any hw-problems in syslog or the dmesg output.
And the OS has not crashed a single time, but Sybase has - on numerous occations - however, without any suspicious messages in syslog/dmesg.
I shall be making arrangements for having the DIMM with the multi-bit error replaced, in order to be able to rule out hw-errors.
If that does not solve the problem, I will return in another thread!

Thanks again,

John K.

[john korterman]

thanks!