By John Burke
Instead of my usual Test Drive, I will present highlights, and perhaps a few lowlights, this month and next from some of the sessions at IPROF '97. A tremendous amount of information was packed into four days, so there is no way I will be able to report on everything of interest (I have a 2-inch binder bursting with notes and copies of slides to prove it). And since my bias runs toward software and management issues, I am ignoring the hardware related announcements and presentations. (Ed note: See the March 1997 FlashPaper for a summary of these) If you have any questions about these or other items from IPROF, contact me directly or Ron Seybold and we will try to track down the answers.
After a multi-year absence, Scott Hirsh is again at the helm of SIGSYSMAN and, with the agreement of those in attendance, is planning to steer the group back to a more HP 3000-centric focus. Scott also plans to reinstitute quarterly meetings of SIGSYSMAN, which will be held in the Bay area. The topic for the first meeting will be password diagnostics (see below).
TURBOIMAGE SCALABILITY. Tien Yu Chen (HP IMAGE lab) gave a presentation on the scalability improvements to TurboIMAGE. Current enhancements are an effort to ensure the performance keeps up with hardware advances. A significant limiting factor in TurboIMAGE is the put/delete semaphore that "locks" the entire database and thus serializes DBPUT, DBDELETE and critical item DBUPDATE. Note: this is different from the programmatic DBLOCK and is not under the user's control. Reasons for this semaphore include:
MULTIPLE JOB QUEUES. Every system manager has struggled against MPE's limitation of one job queue for batch jobs. With the LIMIT command you can control the number of jobs executing simultaneously, but you have little control over the order in which jobs run or the way jobs execute in parallel. Something our Big Iron counterparts have been quick to point out.
This has been a concern of SIGSYSMAN and SIGMPE for years. There have been all sorts of workarounds, third-party solutions and CSL routines devised, but now we will have an MPE solution -- Multiple Job Queues.
Gopi (M. Gopalakrishnan, CSY R&D, Bangalore) gave a presentation on the status of this project. CSY has been working with selected customers on feature sets for some time. However, I believe this is one of the first, if not the very first, significant project where the specifications have been hashed out on the HP 3000-L Internet list. While it may have delayed delivery, the sense of empowerment this process gives to ordinary users is well worth the delay. There are other CSY R&D projects (system-wide CI variables and date intrinsics, for example) in the pipe line that are following this design model. From one ordinary user to HP CSY: thank you for including us.
Job queues will be managed with two new commands, NEWJOBQ and PURGEJOBQ, and with the enhanced LIMIT command. The job queue can be specified in the STREAM command or on the !JOB line. The queue specified in the STREAM command overrides any queue specified on the !JOB line. All jobs go to the system job queue by default. The job queue assigned to any waiting or executing job can be altered with the enhanced ALTJOB command.
The Multiple Job Queues enhancement has a planned beta release of May or June 1997 and is targeted for inclusion in MPE/iX 6.0, which, depending upon whom you listen to, should be released in late 1997 or early 1998.
GREATER THAN 4Gb FILES. It was announced during the Strategic Directions Video Conference some 64-bit functionality will be provided in MPE where it is needed, but that the operating system willnot offer complete 64-bit functionality. Providing for greater than 4 Gb files is an important part of that strategy.
A detailed briefing was given to SIGSOFTVEND on greater than 4Gb files during the closed portion of their meeting (nondisclosure agreements were required). The rest of us were given a peek at HP's strategy (targeted for MPE/iX 7.0) during a presentation by Will McCullough to SIGSYSMAN on what HP is doing to increase or remove various system limits. Other topics discussed included increasing the number of DTCs supported from 128 to 400; increasing the number of FDDI cards supported to greater than one; increasing the number of TCP/IP connections to greater than 6,000; and increasing the number of disk drives supported to greater than 255).
What we know about greater than 4-Gb files so far is that HP is planning to use a mechanism similar to that used to provide jumbo datasets: apparently, a string of less than 4-Gb files will be linked together transparent to the user. Of course we can create a greater than 4-Gb "file" today by linking smaller files and managing the links ourselves. The key is whether HP's solution subsystems such as SORT can handle these "jumbo" files. I intend to keep an eye on this topic. As further information is made available, we will report on it in The 3000 NewsWire.
PASSWORDED DIAGNOSTICS. Password protected diagnostics were discussed at length. But first, a little background. With the introduction of MPE/iX 5.0, Hewlett Packard began protecting various SYSDIAG diagnostic tools with passwords. HP's stated rational is twofold: protect its intellectual property from use by would-be competitors for hardware maintenance and save the user from accidentally shooting himself in the foot. Though one suspects it is more the former than the latter, if you have a current hardware contract, you are entitled to the password. However, there is considerable confusion about how to obtain it. Even worse, if you do not have it and you really need it, like on a Sunday at 2 AM there is no guarantee you will be able to immediately obtain a valid password.
It is extremely unlikely HP will reverse the decision to password protect certain diagnostics (Ed. note: the same password scheme applies to HP 9000 systems). SIGSYSMAN chairman Scott Hirsh reviewed the current procedure and brainstormed with the SIG members on alternative delivery methods. The best solution is to use us support.hp.com, because this site already requires registration with a valid contract for access to HP's knowledge base and thus has a mechanism in place that could be used to supply the diagnostics' password. Scott accepted the challenge to work with HP to see if this could be implemented. The current procedure is outlined below:
1. Call HP License Administration 800.538.1733 and provide the administrator with your contract number or system handle.
2. HP will fax you the license agreement for you to sign and fax back, after which HP will fax you the current password. Your name and fax number will be added to the database to receive the new password every six months.
Even if you have never needed a diagnostic that was password
protected, I strongly
suggest you get yourself in HP's database to receive the current password.
As a system
manager, rule number one should be hope for the best, but plan for the
SIGMPE ENHANCEMENT SURVEY. One of SIGMPE's responsibilities is to discuss possible enhancements to MPE/iX and prioritize them so as to include the high priority items on the System Improvement Committee Ballot distributed to all members of Interex.
Items from the 1995 and 1996 surveys and the current status of the enhancements (presented by SIGMPE chairman Tony Furnivall) are shown below:
|Independent multiple job queues (1)||Testing
|Complete set of date utilities/intrinsics (2)||
|New job type SECUREJOB (5)|| No plan
|Console over LAN (6)||5.0 Express 3
|Improved NEWGROUP/NEWACCT on user volumes (8)||UDC on
|RECALL command for persistent console messages (9)||No plan
|Provide CI access to file system (10)|| No plan
|Enhanced ALLOW command (1)|| Design
|SHOWOUT to show % complete (2)||No plan
|System-wide CI variables (3)|| Coding
|Greater than 4Gb files (4)|| Design
|Improved inter-operability between MPE/iX and POSIX (5)||Investigate
|Disk mirroring of system volume set 6)|| No plan
|DISALLOW command to disable MPE commands (7)|| No
|HIPRI parameter for ALTJOB (8)||Planned
|STREAM intrinsic (9)||Design
|MPE/iX on workstation hardware (10)||No plan
DISK MIRRORING. Disk mirroring is a software product that allows you to pair off disks in such a way that the data from one is "mirrored" on the other. If you have a catastrophic failure of one drive, the other automatically takes over while its pair is being repaired. The argument in favor of mirroring the system volume set is that without this, disk mirroring is an incomplete solution to high availability and of questionable use overall since there are other ways to deal with the data integrity of user volumes. HP's response has been, and continues to be, that mirroring the system volume set is a high effort project. Instead, HP is pursuing a strategy of recommending disk arrays for the system volume set. It even has a project in the works to eliminate the principle single point of failure in its Model 10 and 20 disk arrays -- the controller. The project involves adding a backup controller with automatic switching if the primary controller fails.
During the "war story" section of another SIG, we heard about propogation of random bit errors, a problem with disk mirroring that makes it less desirable as a high availability option than many of us had previously thought. A read request on a mirrored pair is satisfied by the first available member. A write to a mirrored pair occurs to both members of the pair. A comparison of data on the drives is not made in either case. Thus, whether initially occurring at read or write time, it is possible for random bit errors to propagate in such a way that a read can be satisfied with correct data one time and incorrect data another time.
MPE/iX ON WORKSTATION HARDWARE. HP is very resistant to putting MPE/iX on workstation hardware, citing the high cost to develop the drivers and low payback. HP believes it would not sell more than a couple hundred units. HP's assumption is that the principal reason behind the request is for low cost, small footprint, developer machines. It is looking at reducing the minimum MPE/iX user license from eight to four, and perhaps even two.
POSIX SMOOTHING. The Posix smoothing issues fared much better. Posix smoothing generally means making the Posix environment on the HP 3000 inter-operate better with the traditional MPE environment. Jeff Vance (CSY R&D) led a lively discussion covering the major items on the Posix smoothing ballot which carried over into the Birds of a Feather Unix session. Birds of a Feather Unix evolved into a new SIG, SIGPOS/iX, to be chaired by Mike Hensley of Lund Performance Solutions.
SUPPORT FOR GNU. The number one item on the Posix smoothing survey was support for the GNU tools, which include g++. Obviously, customers are reluctant to use these tools or any applications derived from them, such as the HP ballyhooed port of JAVA to the HP 3000, unless they are confident the GNU environment is supported by HP. Where the actual support comes from, and even how much it might cost, was of less concern to the group than a statement from HP that GNU was of strategic importance to the HP 3000, would be delivered by HP and would be kept up to date. HP representatives appeared to believe such a statement of support was possible.
DISTRIBUTION METHOD FOR FREEWARE. The second item on the Posix smoothing survey was a request for HP to distribute GNU, Java and other freeware via FOS as part of the update and Powerpatch process. This would automatically give credibility to the tools, especially those HP is touting as strategic such as Java and Samba. Currently these toools are mostly available on the CSY Jazz Web server. It would also put the tools into the hands of a much wider audience, since many people do not have the high speed Internet access needed to download multi-megabyte files.
A step in this direction was the MPE FREEWARE tape provided to all attendees of IPROF courtesy of Lund Performance Solutions. Special kudos go to Mike Hensley who tested and packaged the items and also did all the tape duplication. (See related story in NewsWire Bri efs about how to order your copy.) Mike volunteered to maintain the freeware if HP would take over the distribution. Ideally this would be accomplished through a FREEWARE account in FOS that would be much like TELESUP, except there would be a mechanism to allow the customer to suppress loading of the FREEWARE account. I believe this become a reality.
MPE/iX 5.5 EXPRESS 2 CI ENHANCEMENTS. MPE/iX 5.5 Express 2 is scheduled for release in this month or next. The complete Express 2 Communicator article on CI Enhancements, written by Jeff Vance and Gopi is available on the CSY Jazz website. A summary was presented during one of the SIGMPE sessions. Many of these enhancements were triggered by, and designed through, discussions on 3000-L. Of particular note are the enhancements to the LISTF, LISTFILE and PAUSE commands. Users can now discover who is accessing a file, including the remote accessors with LISTF and LISTFILE formats 8 and 9. LISTFILE output can also now be filtered by access type or file code. And the PAUSE command can now cause a wait to occur until one or more jobs reach a certain state. Previously, PAUSE could only cause a wait for a specified number of seconds. This enhancement is particularly helpful in managing complex job interdependencies.
NEW DATE INTRINSICS. SIGMPE devoted two sessions to reviewing and finalizing specifications for the new date intrinsics. These new intrinsics are part of HP's strategy to provide customers with the tools to deal with the Year 2000 problem. Gopi) presented the current specifications, already discussed extensively on 3000-L and put them to the floor for discussion during the first session. He followed up at the second session where the design was "finalized". The specifications call for three new supported date types to go with the current 14 commonly in use (a 4-byte integer representing YYYYMMDD, an 8-byte ASCII representing YYYYMMDD and a 4-byte extension to the current 16 bit CALENDAR format) and seven new intrinsics -- HPCALENDAR, HPDATECONVERT, HPDATEFORMAT, HPDATEDIFF, HPDATEOFFSET, HPFMTCALENDAR and HPDATEVALIDATE. For example, HPDATEDIFF determines the number of days separating two dates and HPDATEOFFSET adds or subtracts a specified number of days, the offset, to or from a given date. I believe these new intrinsics are scheduled for an Express release after MPE/iX 6.0.