CUA buffer swaps in ST02 suddenly increase after 720 kernel upgrade

[Former Member]

Hi

We are running SAP Netweaver 701 (EHP 4 for Application component) and have just upgraded our kernel to 720 patch 223.

The CUA buffer is suddenly swapping massively, and free space very low. It has never done this before...

CUA parameter is set as follows:

rsdb/cua/buffersize 100000 kB

We could of course try increasing this but to what size? And the other issue is that we can only restart this system (major Financial / business SAP system) every 2 months...

What I'm more interested in is, has anyone else experienced this? Is it the kernel upgrade that has caused this, or could it be something else?

Regards

Ross

[Former Member · Accepted solution]

We raised an OSS note and now have a reply from SAP:

Dear customer,

this message has arrived in the ABAP development support.

With kernel 7.20, synchronisation requests concerning the CUA
buffer lead to invalidations of the objects concerned rather than
deletions as was the case in the 7.0x Kernel. Though this was an
(unintended) side-effect of a correction in the 720 Kernel, it should
not negatively impact the system. Even if the buffer fills up with such
"invalid"entries, it's no problem, since these invalid objects will
automatically be displaced whenever space for new objects is needed. So
the difference is that an object that is to be invalidated due to a
buffer synchronisation requets will not be removed immediately rather itwill remain in the buffer until displacement will kick it out.
Nethertheless, there will be a kernel patch to return to the old
behaviour. The related note has number 1740934 (still being processed bySAP) and will be released soon.

Sincerely yours
Immo-Gert Birn
Development Support DDIC and Database Interface

Kernel bug!!!

[Former Member]

Hi Ross,

There are two primary causes for a buffer overflow:

1. Directory overflow
A new object is to be placed in the buffer but there are no free directories. In this case, the number of directories needs to be increased.

2. Overflow in the data area:
There are no free data frames (or extents) available. The buffer size needs to be increased.

CUA buffer setting :

a) Size in the shared memory:
rsdb/cua/buffersize [in kilobytes]
Default: 3,000 [kB]
Maximum: 2,097.148 [kB] for kernel Release 3.I to 6.20 with a patch number lower than 1129. Size restrictions no longer apply to releases above this level.

b) The number of directories is calculated from the buffer size and cannot/must not be set explicitly using a profile parameter

Thanks,

Ravi

[Former Member]

Thanks for replying but I fear you have misunderstood the question, I understand all that, the question is WHY has it suddenly increased, could it be down to the 720 kernel upgrade?

100000 kB is already a large size...