Special Tuning
We recently did a complete redesign of a large portion of our database. The time for conversion from the old to the new is coming soon. Obviously the major part of this conversion, beyond our applications, is the conversion of the data from one schema to another. We've written several conversion programs to handle this. These programs are obviously very I/O intensive as they are reading and writing large amounts of data. We are an OLTP shop and our database is highly tuned for that. I'll have the system to myself on conversion weekend so I'd like to tune the database to maximize the throughput of the I/O. Any recommendations? We are on IDS 7.3 by the way and I've got lots of disk to work with. -------------------------------------------------------- Name: Bill Weaver E-mail: Bill Weaver <billw@fscorp.com> Date: 04/28/99 Time: 09:30:31 (Retrospectively realizes there is no future in hindsight) --------------------------------------------------------
Bill Weaver schrieb: > > We recently did a complete redesign of a large portion of our database. The time for conversion > from the old to the new is coming soon. Obviously the major part of this conversion, beyond our > applications, is the conversion of the data from one schema to another. We've written several > conversion programs to handle this. These programs are obviously very I/O intensive as they are > reading and writing large amounts of data. We are an OLTP shop and our database is highly tuned > for that. I'll have the system to myself on conversion weekend so I'd like to tune the database > to maximize the throughput of the I/O. Any recommendations? We are on IDS 7.3 by the way and > I've got lots of disk to work with. Some thoughts: Your conversion-sessions won't need as much virtual-shared-memory as hundreds of OLTP-sessions, so it may be reasonable to minimize your virtual segment and enlarge your resident segment, especially BUFFERS and locks (beware of long transactions). Checkpoint duration is an issue in OLTP, but on that weekend you will do batch-processing: LRU_MAX_DIRTY -> 60-80 LRU_MIN_DIRTY -> 50-70. Increase FET_BUF_SIZE for your batch-processes, if they are doing lots of inserts. Another thing might be, to fasten index-builds by tuning PSORT_ variables; You may find advice on this by searching "PSORT" in cdi at dejanews. hth martin