Processes distrubution across CPUs

[Former Member]

Hello all,

We have some performance issue and sicenario is explained below.

- We are doing data loads to our production system ( LSMW thingy)

- We had complains that loading is slow.

Config : We are running on ECC 6.0, Windows 2003 cluster server, SQL 2005, 16 Processers and 32 GB RAM on each node of cluster. One server runs database + dialog instance (or additional application instance), other one Primary Application server. We have allocated 32GB swap space on each node.

MSSQL is allocated with 12GB memory setting.

During dataload hours, we have following observation.

i) In ST03N, we can see the delay is due to high CPU time.

ii) in ST06, we can see page in is going up to 16GB during that hour.

iii) From ST06, we can see CPU user utilization is 3 % and system utilization is 3% which is stange.

Possible reason for this problem is all the CPUs are not being utilised ? How it make processes distributed across avalible CPUs.

There be question about the no of Data files ( SQL), we have only configured 4 datafiles. But in this instance we do not think this can be problem as DB time is not high.

Regards,

Krishna

[markus_doehr2]

> Possible reason for this problem is all the CPUs are not being utilised ? How it make processes distributed across avalible CPUs.

You can't.

ABAP is a single threaded runtime and a program running only on one workprocess is only as fast as your CPU (one CPU) is.

If you have multicore CPUs there comes another problem with it: bus speed. Two (or four) QuadCore CPUs are slower (seen in total) as four (or eight) dual or single core CPUs. Intel platforms suffer from a (compared to the CPU speed) slow front side bus which slows down the whole processing speed. This may change with the new Nehalem CPUs that use a different technology to connect CPU and memory.

Did you check using SE30 where the time is getting lost? There may be some inefficient ABAP statements that use too much time.

Markus

[Former Member]

Hi Markus,

It is indeed very helpful piece of information ( as usual from you). Now, going a bit further,

we have done same data load on another server with 2 Physical processors, each quad core ( Total 8 cores) , which is relatively going much faster than our main server 8 physical processors, each dual core ( total 16 cores)

But, I have to mention key info that, again, earlier server got just local disks and later one( main one I have trouble with) got disks on SAN.

Well, true that I/O on SAN connected Prod server is a bit slower than earlier one.

Other than, once I meantioned, is there any area you would like me to through light at ?

Regards,

Krishna

[markus_doehr2]

> Other than, once I meantioned, is there any area you would like me to through light at ?

I would run the specific program through transaction SE30 which will show you in detail where most of the time is spent (database or ABAP). Then one could dig deeper into that specific area. We sometimes have such kind of problems too and this approached proved to be nice for getting an overview.

What you also could do is to split up the LSMW parts to be imported into smaller pieces and run the import in parallel.

Markus

[Former Member]

We have allocated 32GB swap space on each node.

You may want to increase swap space 2 to 3 times of your RAM.

we have only configured 4 datafiles

How much is the data and are these 4 files are on the same disk?

[Former Member]

Manoj Thanks for quick reply.

- Error in mentioning Swap size: We have 60 GB page file ( swap size) on each node.

- Currently size of each datafile is 48GB. Total size of used database is 110 GB + some 38 GB index

[Former Member]

and they are all on same Volume on Same SAN. Not sure if they are on same disk.