A commencement address to those ready to matriculate to faster HP 3000 systems
By Mike Hornsby
Speed costs money. How fast do you want to go? You want to get all of the speed out of your current HP 3000 configuration. There are ways of trimming existing costs and applying the savings to improved performance. This article provides insights and methodologies for saving on ongoing costs and improving interactive response times and batch throughput.
If you’re using DTCs, are the terminals running at 19.2 kbaud? Do you have more than four disks per HP-IB or single ended SCSI interface? Do you have a single LANIC for all network traffic? Have you switched to Jetdirect or LPD-based printing sharing?
Maximize your memory configuration/interleave. Switch to Web or socket interfaces to reduce ‘hpusercount,’ allowing a faster server with a smaller ‘hpuserlimit’. Add disc drives in user volumes to spread transaction management. Add or upgrade tape devices to speed up backup operations. Split your fourth-generation language and reporting licenses into developers and runtime licenses, and put the developers on a separate, limited-user system.
Make sure that datasets are blocked efficiently to save disk space and maximize serial prefetching. If you’re performing chained reads, repack on a periodic basis. Place your heaviest accessed files on Fast/Wide SCSI disks. Use HWMPUT or a serial repack to avoid a deleted entry chain. Split your automatic master for small detail from those associated with large details. Watch for master capacities that cause clustering. Avoid the use of master dynamic expansion except for emergency overflows.
Tune your CQ=152,200,200,2000 if you’re using VPlus. Otherwise, tune CQ=152,200,100,2000. Watch for socket applications such as ODBC always running at 152. Use NSCONTROL SERVER=MIN,MAX to pre allocate server processes.
If you’re using mirroring or RAID 5, reevaluate the requirement for logging. If you are logging, make sure that proper procedures are followed after system aborts.
Object Code Translate and allocate Compatibility Mode programs such as FCOPY and any other user CM programs. Install the Native Mode versions of QUERY and QUAD if you use them.
If you’re using spooling utilities, minimize the number of spoolfiles on the system. Use set stdlist=delete in trivial batch jobs. If a program produces a large stdlist, send it to a circular file. At reboot, the system must recover any spoolfiles that were opened at the time of a crash. This used to be done prior to SYSSTART, but now is a background process — but it still can hog the system for a while. Also, doing a print ;start=-20 on a large spoolfile causes all records to be read, because spoolfiles have variable records.
The backup is usually the single longest, most intensive job. Many times, slow response during the day can be attributed to an online restore. Use the store listing or BIGFILES to identify the largest files. It is a shame to waste time backing up memory dumps or month-end work files in each full back up. It is a crime to wait for these files to roll in during a re-install!
Some tools that are available for free: At www.beechglen.com: BIGFILES, a utility to list the files on the system in descending order by size w/cutoff. QPLUS, a command file script that approximates GLANCE or SOS.
At www.allegro.com: FILERPT, a utility that summarizes file usage from system log files. SYSLOG, a utility that allows dynamic switching of system log file events. RAMUSAGE, a utility that describes memory content better than GLANCE. DBLOADNG, a utility to report on database efficiencies
Don’t use PIC 9 fields in COBOL for arithmetic operations. The overhead for ASCII-to-binary is substantial. Do use VPlus forms caching. This is easy to implement in the COM area, and most terminals and emulators support it.
Use DBUPDATE in place of delete/put; this CIUPDATE feature can dramatically reduce the overhead to modify a search or sort item. Do use mode 6 DBGET w/date cut off. Many programs read down the chain selecting for date >, until end of chain. It is usually much faster to read up the chain stopping at date <.
IMAGE does not prefetch for either mode 5 or 6 DBGETS.
Do use Robelle’s Suprtool to extract/sort work files. Many systems have one or two ‘killer’ batch reports. These usually sequentially read a master and chain to a detail producing a work file that is then sorted and passed to a report or output section. Suprtool can usually produce the same work file in one-tenth of the time.
Do use IMAGE b-trees to replace KSAM look up files. IMAGE now has the capability to incorporate b-tree lookups. It is very easy to implement and very transparent to the application code.
Other cost-saving options
Have leased-line and ISP agreements re-quoted annually. Just by asking you can usually save 10-15 percent. Review hardware and software maintenance agreements annually. Look for items that you already have spares for — terminals, DTCs, and tape drives, or thosethat can be replaced inexpensively.
Eliminate OpenView. Many sites have this product to allow DTC switching. This was bundled into MPE/iX 5.5 as host-based DTC control, along with host based Telnet. Look to grow operations and user staff into programmers. Which would you rather have: a great programmer that doesn’t know the business, or a fair programmer that already knows the business and people well?
Mike Hornsby’s company Beechglen Consulting (513.922.0509) provides contract administration, contract programming, and customized system and application software support, specializing in HP 3000s and promoting an “Anyone can call about anything” philosophy. A former HP senior systems engineer, he co-founded Beechglen in 1988.