Previous Epsiodes
- 0.001 - [Destroyed by space pirates]
- 0.002 - Community Day and Onward
- 0.003 - Revenge of The Space Monster
- 0.004 - Children of Murphy
- 0.005 - Journey to the Edge of Quality
- 0.006 - The Calm Before the Storm of Quality
- 0.007 - Passage to Production
Last Epsiode
- 1.000 - The Final Chapter
The Saga Continues, but not much longer
We've spent years putting off the Unicode conversions, and about a year planning and implementing 4 production conversions, as well as 8 dev/test and a few others. Now that time is short, I wanted to detail my final preparations, and share what I'll be doing while the teams execute this weekend.
I want to see performance and trace information, so I've been busy extracting data from the R/3 anc SCM systems. We learned in the first trials that data structures such as STAT beneath ST03 are altered by the Unicode conversion, so capturing data ahead of time is the safest way to verify performance has not degraded.
I looked at overall transaction statistics, such as dialog response time:
| Task Type | # Steps | Ø Time | Ø CPU Time | Ø DB Time |
| Dialog | 16,470,001 | 701.3 | 113.5 | 383.3 |
I recorded batch run time:
| Report/Transaction | Background Job | Number of Steps | T Response Time (s) | Ø Response Time (ms) |
| RSARCHD | BC-GLB-PPORDERDEL-W-C11 | 4 | 261,146 | 65,286,590.30 |
I've captured RFC workload statistics between systems (see my blog on measuring SCM impact on R/3).
I pulled down values from ST10 (table buffers; see this blog also). I've copied these into a database to calculate invalidation ratios as suggested in Schneider's book SAP Performance Optimization.
Example current results are:
| PUF | TABLE_NAME | BUFF_STATE | INV | BUF_SIZE | SIZE_MAX | ABAP_REQS | INV_RATE |
| ful | AQLSCAT | absent | 13 | 0 | 372840 | 266 | 4.89 |
I grabbed database sizes from DB02, although this is not as easy to save in 4.7.
I have history from DB02 by both week and month, which I'll chart after the conversion to show impact of the full export/import. Recent history shows we've freed up several hundred GB in a last minute archiving push to help shorten the down time.
I also looked at ST02, buffer statistics, to verify that we've expanded storage for moving from 8 to 16 bit character sizes on the application server.
Comparing the non-Unicode R/3 to the Unicode R/3 reveals a better generic table buffer hit ratio for non-Unicode, as we've increased the buffer space to 250MB prior to Unicode. I'll see what happens next week to this performance metric. 
The surprise was poor table buffer performance on our bigger app server. I've recommended we increase this size from 250MB to 300MB, as this server is doing more work and seems to need more data on board.
During the conversions I'll be monitoring performance, mainly for posterity as we think our throughput projections will meet requirements. After the systems are converted and backed up, I'll be verifying that the SAP instance and database are recording new metrics, that SQL performance traces work, and that no surprise program failures are occuring due to resourc constraints. And if all goes well, I'll also be celebrating our Scout troop's Klondkie sled adventures.
[All opinions expressed are mine, not my employer's]
