cancel
Showing results for 
Search instead for 
Did you mean: 

Performance tune

Former Member
0 Kudos

Hi team, I'am facing problems with system performance:

My enviorment is:

->Cluster with Intel Xeon 2.80Ghz

->5,50GB RAM

Running Windows Server 2003, and SQL 2005, with ECC5

The Hitratio is fine with values between 93% and 99%, but in the program buffer i´am facing some problems:

The program buffer as 512.000KB alocated , but in 5 hours of activity the free space changes to 1.092Kb with 0,22% , and

the swaps start do grow to 15.0000, the free directory as 115.352 entries / 90,12%

I also get swaps to in the buffer initial records, and in generic key.

In transaction RZ20 the system has the following Alerts:

21.04.2010 11:33:50 PRD SAP_PRD_00 Program Swap ACTIVE 113,300000 /min > 100,000000 /min (15 Min.) SAP buffer swap exceeds threshold

21.04.2010 10:48:50 PRD SAP_PRD_00 Program Swap ACTIVE 207,800000 /min > 100,000000 /min (15 Min.) SAP buffer swap exceeds threshold

21.04.2010 10:33:50 PRD SAP_PRD_00 Program Swap ACTIVE 82,100000 /min > 50,000000 /min (15 Min.) SAP buffer swap exceeds threshold

21.04.2010 09:48:50 PRD SAP_PRD_00 Program SpaceUsed ACTIVE 100 % > 98 % (15 Min.) SAP buffer:buffer storage usage exceeds threshold

21.04.2010 09:18:50 PRD SAP_PRD_00 Program SpaceUsed ACTIVE 92 % > 90 % (15 Min.) SAP buffer:buffer storage usage exceeds threshold

I tried to change the program buffer to 600.000 but the sytem don´t start.

Thanks & Regards,

NG

Edited by: Nuno Gonçalves on Apr 21, 2010 1:55 PM

Accepted Solutions (1)

Accepted Solutions (1)

markus_doehr2
Active Contributor
0 Kudos

I tried to change the program buffer to 600.000 but the sytem don´t start.

Is this 32bit?

What error message do you get in the workprocess traces when you try to start the system?

Markus

Former Member
0 Kudos

Hi, Markus

Yes it´s 32bit.

The message in sm21 after restart with the program alocation with 600.00 is

-

-


Operating system call &5&5&4 failed (error no. &>E5)

The specified operating system call was returned with an error.

For communication calls (receive, send, etc) often the cause of errors

are network problems.

It could also be a configuration problem at operating system level.

(file cannot be opened, no space in the file system etc.).

-

-


markus_doehr2
Active Contributor
0 Kudos

Yes it´s 32bit.

In that case there's not much you can do.

A process on 32bit can allocate a max. of 2 GB (3 GB) of memory, no matter how much memory you have in your system.

All your buffers together plus the necessary extended memory to work in are more than that value and hence the system does not come up.

I highly suggest switching to a 64bit operating system, 32bit systems are not recommended for production use.

Markus

former_member204746
Active Contributor
0 Kudos

have you configured /3GB /PAE in c:\boot.ini ?

have you configured PHYS_MEMSIZE as per SAP note 88416?

have you considered upgrading to 64-bit(software and hardware)?

Former Member
0 Kudos

Hy Eric,

Yes, the configuration, is like you describe in the OS, the instance parameter PHYS_MEMSIZE

has the value 1500.

former_member204746
Active Contributor
0 Kudos

so, this leaves one solution:

have you considered upgrading to 64-bit(software and hardware)?

Former Member
0 Kudos

Hello - Some additional detail here as the limitation regarding memory stems from the 32-bit hardware restrictions (not experienced in 64-bit)

Regardless of 32-bit or 640bit usage SAP note 88416 should be in place.

Zero administration memory management (Note #88416) generally

over allocates roll and page memory. Usually you can reduce the

the amounts allocated to the "in memory"

rdisp/ROLL_MAXFS....Total amount of Roll (Filesystem+SHM); in 8KB block

rdisp/ROLL_SHM......Shared memory element of Roll; in 8KB blocks

rdisp/PG_MAXFS......Total amount of Page (Filesystem+SHM); in 8KB block

rdisp/PG_SHM........Shared memory element of Page; in 8KB blocks

=>Filesystem amount = MAXFS - SHM

60MB or so each in memory should be sufficient. rdisp/ROLL_SHM &

rdisp/PG_SHM only to be changed. This should allow extra

memory space for process memory.

These profiles would have to have be added to the instance profiles.

You should monitor the roll and paging usage after this to ensure that

more than 80% of each is never used as starting to use the paging

and roll files very regularly when buffers are exhausted would have

performance implications. You can check the history of each

instance's usage of these areas by double-clicking on the values from

the initial screen of ST02. This could potentially increase

the available address space per process by about 260MB that would

otherwise be wasted that will help avoid error with the buffer

creation and otherwise allow more memory to be available for HEAP

usage for larger memory intensive reports. These changes do not

invalidate Zero Adminstration Memory Management usage but are merely

tweaking the defaults further.

Again, these are only adjustments and 32-bit limitations still exist obviously.

931024 FAQ: 64-bit platforms for Windows (explains that a 64-bit instance added would avoid issues)

Regards.

Answers (1)

Answers (1)

Former Member
0 Kudos

Hi,

Please revert back to buffer old parameters.

Raise a message with SAP. Is it a new server, did you had Post Go-Live check.

Cheers

Deepanshu