cache feeder is buildet in and working really good.
Thanks a lot for your hints.
Wish you and the whole developer team nice x-mas and a good new year.
with best regards.
Albert
Zabach, Elke schrieb:
Albert Steckenborn wrote:
Hi Elke and Alexander,
now I've slept a little ;-), and then read your messages again. And now I understood. :-)
It's a good idea to feed the cache by doing some standart selects (one per region and Tour Operator like the index is build).
I think this will work and did not need to much time ( in relation to converting and loading).
Is it correct, that the data-cache is held in the main memory?
YES, as all caches of SAP DB / MaxDB are.
Then I've to run the selects at the Main Server before switching online again.
Online in the sense of letting your users work on this server, but of course after online-setting (restarting/doing db_online in dbmcli) of the server.
Elke SAP Labs Berlin
with best rgds.
Albert
Zabach, Elke schrieb:
Albert Steckenborn wrote:application
Hi Alexander,
all our Applications are web-based, so there is no other
data thatthat could feed the cache.
But I think when dropping table and indexes, the cache for
other build inbelongs to that table is deleted too.
So did it makes any scence to delete only the data of the table and not the table itsselve for holding the cached data and all
optimising strategies.
Our strategie with deleting tables is justified in the past with ADABAS D as Database, 20% of datavolume, other machines with single processor, without raw devices and lower ram. Today a load of 5GB data needs about 15min, in the past 2 hours.
I think its time to change the loading strategie to get better performance but I want to make sure that changing will bring up the wanted results (there are a lot of scripts to change).
rgds.
Albert
On one of our sides there seems to be a misunderstanding.
As far as I understood, the deletion, creation and loading
of one table
is done on one database (let's call it A), then the backup
is done and
recovered on database B. Afterwards NO data is in cache of B and the first select lasts too long.
As Alexander says: one first select will feed the cache. And guys
being able to write and run scripts with such complex
SQL/backup-mixture
will be able to add another SQL-call (a select feeding the cache) to this script (in the same 'application' doing the drop table...)
Your idea that not dropping, but deleting the table should
change anything,
seems to be a misinterpretation. The difference is only,
that the meta-data
of the data (and the root-page for the data) have to be
created anew / need
not be created anew. There is no table-specific cache
anyway. There is
no data in the global cache which can be recycled.
In database A, doing the loading of the table, there WOULD
be data in
cache if no FASTLOAD, but the normal one would be used.
But this will not help database B in any case, only the
(possible) selects
on A would run on a feeded cache. Did you check the speed of a normal load instead of a fastload? Why do you switch back to your main server (different hardware/speed ?) ?
On the other hand: perhaps it is not only me, who does not
really understand,
why the whole table has to be filled from scratch 4 times a day.
Are there so many changes? Why don't you update but uses this
back-and-for of servers? Perhaps some info why this
decision was made
would help.too often.
Elke SAP Labs Berlin
Schroeder, Alexander schrieb:
Hello,
it looks to me like the 1st query did not hit the data
cache, and was punished
with lots of data I/O. All following queries possibly then
feed from the cache
to a certain extend and need not to access the hard disk
reload each dayAs dropping, loading, and backup already takes some time,
and is done
before the server is 'switched on' for the application,
possibly executing
the nasty query before from some client program (e.g.
dbmcli) may just
feed the cache, so that the 1st query from the application
isn't really the 1st
one ...
Alexander Schröder
SAP DB, SAP Labs Berlin
-----Original Message-----
From: Albert Steckenborn [mailto:[EMAIL PROTECTED]
Sent: Monday, December 15, 2003 4:47 PM
To: [EMAIL PROTECTED]
Subject: Tuning Question tables (5GB) with complete
with indexes.for 4 times
Hi folks,
of course somebody could give me a hint for my questions.
We are running some applications with tables up to 5GB with SAPDB 7.4 on Linux.
Each day we are getting the whole data again in partial steps (with size
from 0,1 GB to 1,5 GB) at different times.
At the Moment we are dropping the whole table and indexes
Server andper day and creating everything new, loading new data with fastload to our Standby Database-Server. At this Database-Server we are doing everything else to do (creating indexes a.s.o.). Then we are running a backup of the Database switching the applications to this
No way forrunning a recover of our backup at the Main Database-Server. After this we are switching back to the Main Server.
Now our Problem:
The first selects (over 1.500.000 rows) after this needs up to 30 seconds for getting the results. Thats definitiv to much.
reason ofoptimising indexes (we have done our best. Explain looks nice). With the second select, the result is coming up directly (0 sec. in
caching).
Now the Question:
Do you think it is making a scence for performance issues to drop only the data (and reload the new) and not the whole table
Of course you have an other idea what we can do.
With best rgds.
Albert
--
MaxDB Discussion Mailing List
For list archives: http://lists.mysql.com/maxdb
To unsubscribe:
http://lists.mysql.com/[EMAIL PROTECTED]
--
MaxDB Discussion Mailing List
For list archives: http://lists.mysql.com/maxdb
To unsubscribe:
http://lists.mysql.com/[EMAIL PROTECTED]
--
MaxDB Discussion Mailing List
For list archives: http://lists.mysql.com/maxdb
To unsubscribe:
http://lists.mysql.com/[EMAIL PROTECTED]
-- MaxDB Discussion Mailing List For list archives: http://lists.mysql.com/maxdb To unsubscribe: http://lists.mysql.com/[EMAIL PROTECTED]
