Re: Performance. .oO(not AGAIN!)

In article <322457dd.441457462@news.snafu.de>, Daniel Denes <dad@berlin.snafu.de> writes >Hi there, > >here is another performace problem posting. I've read the FAQ but haven't >actually found something that fits. > >The system in question is an older HP 9000/847 with HP-UX 9.0 and Online 5.0 >running on it. It has 128 MBytes of Memory and a HP SCSI Harddisk containing >Informix' raw partition. The application running on this system is a larger DB >application with about 36 tables, each containing 1 to 1.000.000 rows. There are >several procs of various kinds working on that data; the idea is that people >call the datacenter and take some action on their data. Currently there are >about 64 sqlturbo processes, each one possibly running some instances of sacego >at some time. > >When the data center started in '92 performance was nothing to think about. But >now we get into some trouble. So here are my questions: > >At times of high throughput the operators suffer from response times of about 10 >seconds. Sometimes they get the ISAM Error -154 (Deadlock timeout expired - >possible deadlock). I've looked up the program sources and found something like > >EXECUTE IMMEDIATE SET LOCK MODE TO WAIT 10; > >Did this cause the error or may there really by a deadlock? > This means the process is waiting for > 10 seconds to get a lock. Possibly because of deadlocking or possibly just the machine 'running slowly'. >And, more important: What can be done to improve performance? Many,many things - 1. How many disks does the machine have? - Try spreading the database across several disk. i.e. rootdbs on ooe disk, logical/physical logs on a second disk and the database spread across several other disks. You may need to repartition the disks to do this. 2. sacego sounds like ACE reports - try rewriting them as true 4GL and Informix-4GL Dev. i.e. true C4GL reports. 3. Online 7 may well help if you have >1 CPU. Even with 1 CPU Online 7 may be faster. 4. Run sar to check the free memory and see if any paging/swapping is occuring. You may need to add more memory to the machine. 5. If you have >10% free memory run tbstat -p are the %rcached >95% and %wcached > 85% if not increase the OnLine BUFFERS paramater. >I don't think there is much to do with the code, nor is it a problem of >UPDATE STATISTICS (I've tried that! only marginal improvement). If there was>some money to spend on Hardware, what would you suggest? Can Online stripe the >raw partition on several disks? (I've used glance, a performance monitor. It >shows CPU and disk as bottlenecks). Any ideas?? > >TIA, > >Daniel E-mail me if you need more info. -- David Williams