[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference vmsdev::vmstuning

Title:VMSTUNING
Notice:Welcome to VMSTUNING
Moderator:EVMS::HALLYB
Created:Sat Feb 15 1986
Last Modified:Wed May 14 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:1539
Total number of notes:7984

1539.0. "SRP allocation failures" by OSOV03::KAGEYAMA (Bring us back DECps!) Wed May 07 1997 08:24

I know there's "IRP ALLOCATION FAILURES" topic, 1134, but it does not 
answer my question. So here it goes.

My customer experienced SRP allocation failure while their SRP did not 
exceeed SRPCOUNTV. OpenVMS VAX 5.52-H4 on VAX 4108A.

- MPW_LOLIMIT is 5900, FREELIM is 150, WSMAX is 262100(So 65535 would be
 used), and pageable pages, PFN$GL_PHYPGCNT, are 913750. So SRP should 
 be expanded to SRPCOUNTV.
 
- This is a single processor system.

- Memory utilization is less than 60%.

What kind of thing could cause expansion failures? PMS$GL_XRPFAIL shows
11 times of failure within 53 days.

Small Packet (SRP) Lookaside List            Packets       Bytes       Pages
    Current Total Size                        170144    21778432       42536
    Initial Size (SRPCOUNT)                   100352    12845056       25088
    Maximum Size (SRPCOUNTV)                  504915    64629120      126229
    Free Space                                  8926     1142528
    Space in Use                              161218    20635904
    Packet Size/Upper Bound (SRPSIZE)                        128
    Lower Bound on Allocation                                 32

Best regards,

- Kazunori
T.RTitleUserPersonal
Name		DateLines
1539.1some things to checkCSC32::BUCKLEYski fast,take chances,die youngMon May 12 1997 09:4412
srp's are normally used for locks...
what are the values of lockidtbl, lockidtblmax and the current number of locks
being used (moni lock).
Could this be from a resource tree being remastered to this node.
sda>sh sym/all pms$gl_rm for values related to remastering

did they run out of npagedyn ($sh mem/pool/full) which is where the srp list
goes when it has to expand?

Dan Buckley
Digital CSC
1539.2OSOV03::KAGEYAMAI Got Rhythm!Tue May 13 1997 05:1954
Dan,

Thanks for the reply.

>srp's are normally used for locks...

I should have said 88% of SRP are occupied by unknown packets and this 
part is growing.

>what are the values of lockidtbl, lockidtblmax and the current number of locks
>being used (moni lock).

LOCKIDTBL is 6310 and the lock count was around 16,000. LOCKIDTBL will 
be increased to 20,480.


>Could this be from a resource tree being remastered to this node.
>sda>sh sym/all pms$gl_rm for values related to remastering

PMS$GL_RM_ACQUIRE               = 80004FC0 : 00002E22
PMS$GL_RM_BETTER                = 800051B4 : 00000000
PMS$GL_RM_FINISH                = 80004FC4 : 000069FA
PMS$GL_RM_MORE_ACT              = 800051B8 : 00003BFC
PMS$GL_RM_MSG_RCV               = 80004FD0 : 0000F470
PMS$GL_RM_MSG_SENT              = 80004FCC : 00011DA6
PMS$GL_RM_QUOTA_WAIT            = 80004FB8 : 0000478E
PMS$GL_RM_RBLD_RCVD             = 80004FD8 : 00079056
PMS$GL_RM_RBLD_SENT             = 80004FD4 : 000E6C56
PMS$GL_RM_REQ_NAK               = 80004FC8 : 00000024
PMS$GL_RM_SINGLE                = 800051B0 : 00007595
PMS$GL_RM_UNLOAD                = 80004FBC : 00003BFC
PMS$GL_RM_QUOTA_WAIT            = 80004FB8 : 0000478E
PMS$GL_RM_UNLOAD                = 80004FBC : 00003BFC
PMS$GL_RM_ACQUIRE               = 80004FC0 : 00002E22
PMS$GL_RM_FINISH                = 80004FC4 : 000069FA
PMS$GL_RM_REQ_NAK               = 80004FC8 : 00000024
PMS$GL_RM_MSG_SENT              = 80004FCC : 00011DA6
PMS$GL_RM_MSG_RCV               = 80004FD0 : 0000F470
PMS$GL_RM_RBLD_SENT             = 80004FD4 : 000E6C56
PMS$GL_RM_RBLD_RCVD             = 80004FD8 : 00079056
PMS$GL_RM_SINGLE                = 800051B0 : 00007595
PMS$GL_RM_BETTER                = 800051B4 : 00000000
PMS$GL_RM_MORE_ACT              = 800051B8 : 00003BFC

I don't know how to translate above values.

>did they run out of npagedyn ($sh mem/pool/full) which is where the srp list
>goes when it has to expand?

One third of nonpaged dynamic memory is free.


- Kazunori
1539.3OSOV03::KAGEYAMAI Got Rhythm!Tue May 13 1997 05:238
The back ground of my question is...

My customer experienced performance slow down and I thought it might be 
related to SRP allocation failures which might cause high ratio of 
kernel mode usage.

- Kazunori
1539.4other things to checkCSC32::BUCKLEYski fast,take chances,die youngWed May 14 1997 09:379
Your didn't say how long the machine was up, the highest value from sda was 
900,000 rebuilds sent over the time the system was up.

If this is a v5.5 system there is a possibility that a bug in servicing the 
ipl 6 fork queue is biting you (sorry didn't re-read the first note).
(fixed with patch CSCPAT_0253 or vaxshad09_u2055)
Is the pagefile over 50% full?

The heavy use of spr's could just be a symtom of heavy load.
1539.5The real problem might be negligence of SRP releaseOSOV03::KAGEYAMAI Got Rhythm!Wed May 14 1997 23:0929
>Your didn't say how long the machine was up, the highest value from sda was 
>900,000 rebuilds sent over the time the system was up.

.0>What kind of thing could cause expansion failures? PMS$GL_XRPFAIL shows
.0>11 times of failure within 53 days.

>If this is a v5.5 system there is a possibility that a bug in servicing the 
>ipl 6 fork queue is biting you (sorry didn't re-read the first note).
>(fixed with patch CSCPAT_0253 or vaxshad09_u2055)

.0>exceeed SRPCOUNTV. OpenVMS VAX 5.52-H4 on VAX 4108A.

I'll check the patch.

>Is the pagefile over 50% full?

No, it's less than 15%.

>The heavy use of spr's could just be a symtom of heavy load.

It grows and grows, and UNKWON part of SRP looks like the culprit. This
system is used for ALL-IN-1, Notes and ANETplus, which is a TCP/IP 
communication product predated to UCX in Japan. SRP usage grows in the 
day time and decreases in the evening, but it does not restored to the 
initial stage of the morning. I wonder some SRPs are not released by 
someone, but couldn't trace it now.

- Kazunori