problem with multitape level 0 archive

Hi, I have one for all you On-line experts out there. We have a very large On-line 5.0 application of an extremely critical nature (Enhanced 911 for several US RBOCS and large Independent telcos.) This application runs on a mirrored pair of Starserver FT computers. Here is our problem: at one of our sites, a level 0 archive spans several tapes. If, when a tape is written and On-line is prompting you to install a second tape, and for whatever reason, On-line decides that it is time for a checkpoint, guess what! The server is hung and everything stops until the operator puts in the new tape! On-line will do nothing until it can execute the checkpoint, and it will not execute the checkpoint until a new tape is installed. I thought On-line was supposed to provide "on-line" backups! Are they "on-line" as long as they don't span multiple tapes with an intervening checkpoint? Thats not what the salesman told us! OK, having said that, let me describe the workaround Informix proposed. Basically, our problem is that checkpoints hang the server when backups are running and we force checkpoints very often because we want checkpoints to remain very short. Well there is another way to keep checkpoints very short, that is to set the tuneables LRU_MAX_DIRTY and LRU_MIN_DIRTY very low, like 5 and 0. These tuneables control how much of shared memory can be dirty at any given time. By setting these very low, not very much of memory can be dirty, therefore checkpoints don't take very long. With these changes, we can set the checkpoint interval to something quite long, like an hour, without worrying about checkpoints taking more than a second. Then while doing backups, time how long it take to fill up a tape. About 5 minutes before you expect the tape to finish, force a checkpoint. This will give the user at least 55 minutes to change the tape without anything bad happening. This is definitely a hack job, and has a number of holes in it. Most obviously, it doesn't solve the problem, it just lessens it. If it takes them 2 hours to change the tape, we are just as stuck as we started out being. Second, timing the first tape will be easy, but timing for the breaks at the second and third tapes will be a little more dicey since it will be hard to tell exactly when they do change the tape. Third, a checkpoint will also be forced if the physical log fills up. The physical log is where dirty pages are buffered between checkpoints. Therefore, even with these changes, if the system were very busy, it is possible that checkpoints could occur more frequently that an hour. This could be lessened as well by cranking up the size of the physical log and timing the backups, but again, this does not really solve the problem. If anyone out there at Informix is reading this, please respond and TELL ME NONE OF THIS IS TRUE! Is this fixed in 6.0? 7.0? By the way, slamming Informix is not my intention. We have been extremely pleased with the product, the performance and the tools. On-line's performance is I believe, the best in the industry, and our customers are very pleased. Its just that they have this one small problem when you do a level 0 archive that spans multiple tapes . . . *----------------------------------------------------* | Larry Ciesla - MTS | ciesla@stairs.att.com | | AT&T Network Systems | | | E911 Development | 708-224-6747 office | | 2600 Warrenville Rd | 708-224-4524 fax | | Lisle, IL 60532 | | *----------------------------------------------------*