on 10-01-2012 8:08 PM
Hi SAP
We have just upgraded from SAP ERP 600 EHP4 to ERP 600 EHP6
(NW7.31 SP03, SAP_APPL SP03). OS AIX.
After the upgrade we seems to have an performance -/capacity
issue concerning ICM/http connections ( - WebCost, POWL)
Users seems to loose the http user connection.
TA ST11, Dev_icm_sec_4
Error: Timeout occured (-5), Traffic Control: Nettimeout (30) exceeded
by peer:
TA ST11, Dev_icm
[Thr 2828] Mon Oct 1 12:00:00 2012
[Thr 2828] *** ERROR => no more free event-flags.
[mpixx_mt.c 5787]
[Thr 2828] *** ERROR => MpiIEvtOpen(0) failed: 11
[mpixx_mt.c 1668]
[Thr 2828] *** ERROR => IcmConnInitBuffer: MpiCreate2 failed
(rc=MPI_EEVT: Event flag error/11) [icxxconn_mt. 4282]
[Thr 2828] *** ERROR => no more free event-flags.
[mpixx_mt.c 5787]
[Thr 2828] *** ERROR => MpiIEvtOpen(0) failed: 11
[mpixx_mt.c 1668]
TA ST11, dev_icm
[Thr 26729] Mon Oct 1 12:04:09 2012
[Thr 26729] {00ac137b} Traffic Control: Nettimeout (30)
exceeded by peer: 10.49.59.13:49462 [icxxthr_mt.c 4285]
[Thr 26729] CONNECTION (id=172/70523):
used: 1, type: default, role: Server(1),
stateful: 0
NI_HDL: 2711, protocol: HTTP(1)
local host: 145.218.246.102:1080 ()
remote host: 10.49.59.13:49462 ()
status: READ_REQUEST
connect time: 01.10.2012 12:03:37
MPI
request: <2650e>
MPI response: <2650f>
request_buf_size:
0 response_buf_size: 0
request_buf_used:
0 response_buf_used: 0
request_buf_offset:
0 response_buf_offset: 0
TA ST11, Dev_icf31
</ABAP-CallStack>
</ErrorInfo>
<ErrorInfo URL="http://global-erp.dsv.com:1080/sap/bc/webdynpro/sap/ziwf_powl">
<ErrorMessage>User session (HTTP/SMTP/..) closed after
timeout</ErrorMessage>
<Date>20121001</Date>
<Time>111940</Time>
<Client>700</Client>
<User>DKBJJ</User>
<TerminationType>ERROR_MESSAGE_STATE</TerminationType>
TA SM21
DP Q0 G Request (type DIA) cannot be processed - SAP Web AS Problem ST/Problenm cl
The amount of memory pipes mpi/max_pipes is not set in instance profile, and
as so from documentation should be default value 4000. Somehow this parameter
can’t be looked up in TA RZ11 ?
Minimum event-flags/OS
semaphores for AIX seems as from link
for AIX in note 715400: ” ver 5.3 Maximum
number of semaphore IDs for 64-bit kernel: 1048576 ”. As from link for online
documentation on ”Parameters for Memory Pipes”,: “ The default setting
only has to be changed for very big systems or for the Web Dispatcher that has
to keep open a lot of parallel connections”.
Notes and links that we have looked at,
Note 715400 - Max. number of memory pipes (MPI) restricted to
4,000
“Symptom
Parameters for Memory Pipes
http://help.sap.com/saphelp_nw73ehp1/helpdata/en/48/3f044575d431cbe10000000a42189d/frameset.htm
mpi/max_pipes (default
4000)
Memory Pipes and MPI Buffers
http://help.sap.com/saphelp_nw70ehp2/helpdata/en/48/1eeaa5c26a4e73e10000000a42189c/frameset.htm
“Operating System Resources
Two memory pipes, four event flags, and two sockets are used for
each HTTP connection.
The number of event flags (two for each memory pipe) is restricted
by the operating system. SAP guarantees at least 40000 flags for all supported
platforms. The maximum value for parameter mpi/max_pipes is 20500.”.
Note 737625 - Parameter recommendations for the ICM
Thanks in advance for any reply on this issue.
Best regards
Hanne Münther
Hello Hanne,
The amount of memory pipes mpi/max_pipes is not set in instance profile, and
as so from documentation should be default value 4000. Somehow this parameter
can’t be looked up in TA RZ11 ?
Use RSPARAM report or RSPFPAR transaction/report to find the current value.
Best regards,
Adam
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Hi Adam
THANKS a lot for your fast reply on this issue.
Entries of typen “[Thr 2828] *** ERROR => no more free event-flags. [mpixx_mt.c 5787]” seems to have stopped.
Users still loose their http connections and entries as in Note 1684891 - System log Q0G due to cancelled user sessions shows in TA SM21 and TA ST11, dev_disp.
BUT the users do not as in the note cancle their session manually or close their browser, they seem to get a time out error.
The reason for these time outs, that users experience, seems to be shown in TA ST11, dev_icm as entries ” [Thr 26729] {00a2c23b} Traffic Control: Nettimeout (30) exceeded by peer: 10.49.59.48:49767 [icxxthr_mt.c 4285]”. Note “1629748 - traffic control timeout does not work properly”, mentions these entries in dev_icm, and mention the instance profile parameteren icm/traffic_control . Default value 30 sec, as too shown in entries in dev_icm.
BUT WHY does these entries show ” Traffic Control: Nettimeout (30) exceeded by peer:” ? As from the note and linket from help.sap.com on parameteren icm/traffic_control, then the reason is ” This parameter defines the timeout within a request. If the request is not yet fully transferred, and no data has arrived for some time in this timespan, the connection is closed.”. The default value of 30 sec worked ok before the ERP600 EHP6 upgrade and the system is not heavy loaded.
Should we increase the value of icm/traffic_control ?
Somehow we cannot look “icm/traffic_control” up in TA RZ11, SE38 report RSPARAM report or in TA RSPFPAR. As from http://help.sap.com for NW731 (http://help.sap.com/saphelp_nw73/helpdata/en/97/a6d779d2f343fb8fab5e141b69bce0/content.htm) ., it should be available for NW731.
best regards
Hanne Münther
Hanne
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Just to mention that we have solved our problem by applying note 724713 which is related to Solaris 10 and SAP kernel 720 patchlevel 104 or above. In a nutshell, OS limits like the number of semaphores, shared memory segments, ... are not managed via /etc/system anymore but via dynamic "projects"
Best regards
Luc Eraly
User | Count |
---|---|
91 | |
10 | |
10 | |
9 | |
9 | |
7 | |
6 | |
5 | |
4 | |
4 |
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.