Re: Write performance

Jon Myatt wrote: > > Finger in the air time. > > I'm after some figures for time taken to insert/update 5000 records, > where each row is about 200 bytes, (40 cols) and the table has > a PK and two other indexes. (In my case, the table is fragmented > across 4 disks, indexes on separate disks, logs etc on separate disks, > On-line 7.24). > > I know this is *extremely* vague, but if anyone can give me > some times taken to do something like the above, with details of > the hardware used, I'd be *extremely* grateful. > > Even better - if you have a spare 5 mins and can test this out > for me, I can email you a schema for the exact table I'm testing. > The more results I can get for this the better. A good rule of thumb for inserts from a single threaded source to a simple disk farm is 100 rows per second. With many processes inserting or using advanced ESQL/C techniques (like FETBUFSIZ and the Fetch Array feature) and a faster disk farm on a multiprocessor platform this will climb in to the many thousands per second. Using 20 copies of an optimal utility on a Sun E6000 w/ 20 330MHZ CPUs and a 20 drive RAID10 disk array it would probably take longer to start up the inserters than to actually add the data. At any rate it should not take much longer than a minute. Art S. Kagel