Export Questions - Large DB

Background:
Using exp/imp to migrate database between two UNIX machines (IBM->Sun, etc) or UNIX -> NT.
Database mostly consists of a single large table (30M+ rows).

Solutions so far:
Use "compress on the fly" with pipes for exp/imp. Use indexfile to build tables, import data (no indexes), build indexes. (Used because initial extent size for tables usually needs to be fiddled with or tables being built in a different tablespace that they came from).

Questions?
I read in Deja News that using pipes slows down the process, because the pipe file size is usually only 8K. I haven't really seen this problem. Does this slow down the process?

For speed should I "compress" out of the pipe or 'split' out of the pipe if I have disk space?

• Can I "thread" this process? *** That is: 1) export from source using pipes and split into "chunks" (maybe compressed) 2) Transfer first chunk to target system. 3) cat chunk* through pipe into import command 4) continue exporting, transferring, importing

Problems-
Initially don't know how many chunks (thus filenames) there will be to import. 'cat' could bomb.
Can't use imp with the indexfile option, because it reads through the exportfile too fast (not all chunks are available).

• Is using rsh to transfer the data during export cleaner/faster? *** How "stable" is this process? It could cut down time drastically, if we could multitask the export and import.

Do I loose significant speed using the create table, import data, build indexes process? I allows flexiblity. I can easily modify table and index create command for initial/next extent for the new platform. If it pukes on an index create (not enough temp tblspace), I can drop and restart the index create easily. What am I loosing?

For NT are there similar processes?

Pete Stryjewski
pstryjew_at_worldnet.att.net Received on Tue Feb 02 1999 - 09:58:47 CST