Q: rollback sizing pt 2

From: Vince Laurent <eAddict_at_yahoo.com>
Date: Thu, 14 Aug 2003 10:43:45 -0500
Message-ID: <7cbnjv473ojvabct9t092cnli72n6onjvp@4ax.com>

My PSAPROLL table is currently 5.4G. I have 20 identical rollback segements. I still get A LOT of 1555 errors - esp on batch jobs running at night. Here is what I am seeing right now:

Name	Extends	Gets	HWMSize	 Wraps	Avg Actv  Waits
-----	-------	------	--------- ----	--------- ----
PRS_0	4	88240	19161088  120   2414587	   0
PRS_1	14	93020	29810688  136   2336032    0
PRS_2	20	99058	29810688  147   2332290	   9
PRS_3	16	98388	31940608  145   2583112	  16
PRS_4	144    155675  178905088  284  10352191	  67
PRS_5	9      100595	23420928  144   2261469	   4
PRS_6	10	95611	21291008  138   5616607	   2
PRS_7	46     115365	97968128  167   2179500	  46
PRS_8	10	98889	21291008  145   2913379	  15
PRS_9	16	99217	27680768  138   3380851	   8
PRS_10	18     595968	38330368  150  10095050	   2
PRS_11	15	87944	31940608  125   2195350	   2
PRS_12	14     103694	29810688  146   2362446	  15
PRS_13	12	95758	27680768  132   2159040	   2
PRS_14	15	99031	21291008  149   4712310	   2
PRS_15	11	97991	21291008  150   4222024	   6
PRS_16	0	85402	10641408  117   2590132	   0
PRS_17	8	90411	27680768  126   2541240	   2
PRS_18	8	92401	27680768  125   2319735	   2
PRS_19	9	89203	29810688  110   2555406	   3

Total Waits:            203
Total Gets:             2484902

Ratio of Waits to Gets: 0.00817%

Avg Actv: Avg # of bytes in active extents in the rbs

          measured over time
(the above chart was generated by the FREE tool "orastat")

In order to make the chart fit, I took some columns out. All segments have 505 max extents, inital and next of 10640408, all are ONLINE, and all have an Optimal size of 10485760. (all set up as SAP defaults)

According to SAP Note 185822, the hwmsize should be smaller than initial+(max*extents). That is the case (as that number would be 1,061,158,912).

According to SAP Note 3807, having a initial of 10M and a next of 10M each should be able to grow to 10*505 or 5G...

My question is: Would it be better to
(a) add another datafile to the PSAPROLL table and make it bigger than

    5.4G?,
(b) add more rollback segements,
(c) reduce the number of MAX_EXTENTS,
(d) increase the size of the individual rollbacks,

or a combination of above.

Rollbacks are confusing.... thanks for any help! Vince

---

Come race with us!
www.mgpmrc.org Received on Thu Aug 14 2003 - 10:43:45 CDT