I have done an ADABAS tuning course a couple of years ago. On that course they suggested you should not make the LFP > 1500000 because of the scan mechanisim ADABAS uses to scan the pool. As I said, it was a couple of years ago. We are running V 7.4.3 now. Have they changed the way ADABAS scan the pool? Can we increase the pool size. What is the suggested maximum size (performance wise). Thanks.
I do not know how many years ago your training was. I was looking thru my old records; and found out following:
- with Version 6.1. the UQE was extended and we keep there an anchor to the private Formatids which are then linked together in the LFP
- with Version 6.2. the unused space in the LFP get chained
- with Version 7.4.2. there was a zap (AN742192) which is included in 743 where the limitation of 5 Global Format IDs in the UQE was removed, the GFIDs are now also chained in the LFP
All together from the theory the recommendation to limit the LFP to have a better performance do not longer apply to Adabas Version 7.4.3
Unfortunally I cannot tell you about the pratical side - I only saw some Requests where other customers have a value of 5.000.000 for LFP and it seems to work…
Personally I would recommend that you increase the LFP and whatch your performance.
Thanks Ursula. I will play around with the sizes and monitor the performance.
I did an extensive study of LFP for the 2004 Boston Natural Conference
(Adaruns as a Sunday class), then modified it this year because of the
zap alluded to above.
If you do not have any Global Format Id’s (most likely a non-Natural
shop), then the LFP should be sized until format overwrites are eliminated.
If you have Natural and/or global format id, there still is a tradeoff
between scanning the global format id chain versus the time it takes
to translate a new gfid ifb (this is analagous to the old Adabas Version 4
discussions). For a database with just global format id’s, the formula
for LFP should be (based on cpu studies of scan time versus translate
LFP = (AVER.IFB.LEN) * N.GFID.IFB
N.GFID.IFB = 2 * 30 * (AVER.FLDS/FB)
AVER.FLDS/FB = FBL / 9
average FBL can be obtained by command log samling.
Thanks Jim for the info. We are a pure NATURAL shop.