Calculating p_manage_01 work space

Last updated
Save as PDF
Share
1. Share
2. Tweet
3. Share

Article Type: General
Product: Aleph
Product Version: 19.01

Description:
In the"How To Run Index Jobs", you say: "The most important concern is that the $TMPDIR be in a different file system than the others." I am wondering if this is still of concern, in this age of SAN disk... This is the type of disk we will be using for the parallel indexing, according to the mail from Computing Centre:

220GB Tier 2 storage (AMS2100) virtualized behind the Tier 1 disk array.
Library will get the benefit of Tier 1 virtualization

This "storage" is split into 2 mounts:
69G 12M 68G 1% /loandisk1 - working space
148G 144G 2.1G 99% /loandisk2 - Oracle tables

So my question is, can I redirect both $TMPDIR and NDX01's scratch to loandisk1? I would keep NDX01's FILES under the /exlibris1 mount. (on NewAleph, our U_tree is here, with a symbolic link in /exlibris).
56G 28G 23G 56% /exlibris
56G 17G 33G 35% /exlibris1

I want to make sure I don't fill up /exlibris1...

Resolution:
SAN does not change the undesirability -- from a performance standpoint -- of reading from and writing to the same filesystem at the same time.

The steps in the p_manage_01 index job are:

1. read z00 records from the oradata disk

2. write a work record to $data_scratch

3. read from $data_scratch, write a sort record to $TMPDIR

4. read from $TMPDIR, write to $data_scratch

5. read from $data_scratch, write sequential file to $data_files

6. read from $data_files, write z9n records to Oracle.

Thus, there are five disks involved: the z00 disk, the $data_scratch disk, the $TMPDIR disk, the $data_files disk, and the z9n disk.

The example in section 5.1 of "How To Run Index Jobs" was taken from McGill -- when you had 1.7 million records instead of your current 3.1 million:

mgu01@PALEPH1> select count(*) from z13 where z13_title is not null;
**** Hit return to continue ****

COUNT(*)
----------
3125021

Thus, you have 80% more records than you did then. Multiplying the section 5.1 number for $data_scratch by 1.8, we get 57 gig. This will take up most of /loandisk1 by itself.

If you leave the number of processes at 7, the $TMPDIR space will *not* increase, so that would be 20 gig (probably only 15 gig used).

And likewise $data_files: 10 gig free (probably only 5 gig used).

Given which, I think you should move only the mgu01 $data_scratch to /loandisk1. The 33 gig on /exlibris1 *should* be enough for both $TMPDIR and $data_files -- though I would suggest cleaning /exlibris1/log directory as much as possible before starting.

Article last edited: 10/8/2013