Number 48 (Update of Volume 5, Issue 5)
Welcome to our 48th monthly edition of Online Extra
the e-mail update of our articles in recent issues of the 3000
NewsWire, plus items that have surfaced since we mailed our previous
First Class issue (February, 2000). We e-mail our subscribers this
file between the First Class issues they receive by mail, updating
stories and adding articles that have developed between issues.
Ron Seybold
Editor in Chief
HP corrects performance ratings; benchmarks
still in limbo
After
we ran the first article on the newest high-end systems in our
February issue, HP learned about a mistake in its latest performance
figures for the systems. Figures released about the e3000 performance
ratings got revised by HP, when customer Jeff Woods discovered that
the reported number for a 6-processor Series 997 looked like it went
down from a 4-processor system. The corrected numbers for the Series
997 line, as represented in HP 3000 Performance Units where 1 is a
Series 918:
997/200 -- 13.2
997/400 -- 23.7
997/600 -- 32.2
997/800 -- 39.0
997/1000 -- 48.4
997/1200 -- 52.3
Kevin
Cooper of the e3000 division's performance labs said that the boosts
in the newest models of the line, the /1000 and /1200 10-way and
12-way 997s, are "due to new software features that are only
enabled on 10-way and 12-way systems running MPE/iX 6.5."
When
customers were examining the new performance marks, one noted that
having a benchmark available to compare HP e3000s with other systems
would be useful in getting management buy-in for 3000 upgrades. Some
customers have hoped for a TPC-C benchmark on the HP e3000, something
that the division has been loath to invest its resources in for
several years now. One solution would be to get a group of customers
and consultants working on the project, with help from HP through the
new Shared Source program. Greg Stigers, a consultant at IT services
company CGI, said that "I am hoping that a Shared Source
approach could get the ball rolling, and generate enough interest to
keep it rolling, while limiting the cost to any one person, group, or
company."
Views
of e3000 division (CSY) managers towards a TPC project have been that
benchmarks would come at the expense of other technical projects
within CSY. A typical viewpoint is the one that former lab section
manager Jim Sartain told us about in 1998, in our Q&A interview.
Sartain said, "Giving them a TPC [benchmark] number isnt
really going to be that persuasive. In fact, often those numbers
dont reflect the real world. Theres a big investment to
do TPC benchmarking in hardware and people resources, which has
marginal customer benefit." (Interestingly enough, Sartain is
now working in R&D at the HP Open Skies operation, which is
selling an application to customers who've never used an HP 3000.
Open Skies isn't selling systems or software, though just
transactions in the apps on tap model.)
A
little later on, HP's Solution Provider Program liasion Narinder
Sandhu said that benchmarking the HP e3000 would be most likely to
happen to leverage a big deal in an application sale, and HP could be
counted on to help in that circumstance. But customers still working
on defending the platform to their management say a TPC-C mark would
help keep the 3000 from getting turned off.
Stigers of CGI and Gavin Scott of Allegro Consultants have
both shown interest in the project. Scott said his company would be
"happy to give [CSY] a proposal for a preliminary investigation
into architecting an "MPE" implementation of TPC-C,
including rough performance projections, etc., as a first step to
determine more clearly whether it's something that's both worth doing
and practical from a cost point of view."
Scott
explained that "The way to do TPC-C on a 3000 is to write a
'3000 application' that implements TPC-C using native TurboIMAGE
calls and all the other good MPE programming features. It would
certainly not be a small amount of work, but given a reasonable
number of clever programmers, I think you could get some very
impressive numbers out of MPE/iX."
Suggestions to get TPC benchmarks for the e3000 tend to come
from the installed base customers who are using the system for
home-grown applications, rather than the packaged, off-the-shelf apps
such as Amisys and Smith-Gardner's WebOrder. The home-grown customers
make up a much larger part of the e3000 installed base sales, but the
packaged customers are in the majority of new HP 3000 sites. We can't
think of a better project to show how Shared Source can help the
entire e3000 customer base. While watching HP try to drum up help for
its Java class library Shared Source project, it looks like Shared
Source projects might need an advocate, someone to recruit
volunteers. Since the e3000's advantages include transaction
performance second to none, it's hard to see why a marketing effort
for the system wouldn't benefit from this kind of advocacy.
3kworld does facelift on site, links live to
3000-L listserver
New
navigational aids and better links to external content are the early
plums in a revision of the 3kworld.com Web site this week. The new
design went live on Friday, introducing the concept of
"IntraCommunity Portals (ICPs) -- think GeoCities.com for the
3000 community." The reorganization is expected to highlight
participation by e3000 customers who focus on vertical packages like
eXegySys manufactuing or Mitchell-Humphrey financial software, by
helping those customers see the array of specific chat rooms, message
boards and online surveys connected to those vendors' ICPs.
The
Web site is also increasing its content for general purpose HP 3000
managers, including a live link to the 3000-L mailing list traffic
beginning on Monday the 20th. Content manager Brian Trembath said
3kworld.com visitors will be able to read all messages from the
mailing list through the Web site, and post replies that can be read
by mailing list devotees as well as the Internet newsgroup mirror of
the mailing list. In short, the plan looks to extend the community
exchange of the mailing list to a new set of e3000 sites, those that
don't use the newsgroup or mailing list interface today.
3kworld.com is also providing original content through the
auspices of its parent, e3000 North American distributor Client
Systems. On March 23rd Chris Gauthier, Client Systems Technical
Support Engineer, offered a tutorial on DLT 8000 in MPE/iX.
Administration guidebook for HP e3000 gets
underway
Jon
Diercks, who's contributed articles on the 3000 for the NewsWire in
the past, dropped us a note to report he's working on a new technical
book for MPE/iX administrators. "I wanted to let you know that I
have been contracted by Prentice- Hall/HP Professional Books to write
the MPE/iX System Administrator's Handbook, scheduled for October
release. I am very excited about this project and I'm making good
progress. With luck, we may even have a few advance copies to sell at
HPWorld in September." We hope to publish some excerpts from the
book closer to the publication date.
HP Jazz Web site gets another tutorial
The
Jazz Web site operated by the e3000 division has a wealth of
information to help manage HP 3000s, created by HP engineers and
managers. The site recently gained a new tutorial from the HP 3000
Solutions Symposium. Alvina Nishimoto's talk on middle options and
differences for e3000s is online in an HTML format that should run
well in most browsers. Steer to http://jazz.external.hp.com/papers/SolSymposium_00/middleware
/index.htm to step through the slides from Nishimoto's talk. It's
also online as a PowerPoint slide show.
e3000 fans speak out in InformationWeek
After
HP's work to get its e3000 rebranding noticed by the general computer
trade press, one article in InformationWeek used the phrase
"decades-old, proprietary HP 3000" in describing the
system. Word circulated in the community, and a letter-writing
campaign to set the record straight yielded a big response. Reporter
Martin Garvey at InfoWeek told SIGIMAGE chair Ken Sletten, "We
have never gotten so many letters for any section story, maybe even
news stories." (Whatever his mode of description, Garvey is to
be commended for picking up the e3000 rebranding story, one that got
away from other publications such as Computerworld and PC Week.
JavaWorld's Julie Salzmann ran a brief item on the rebranding in her
Feb. 21 issue as well. You can read the InfoWeek story at http://www.informatio
nweek.com/773/hp.htm).
Two
of the e3000 community's letters appeared in the Feb. 28 issue of
InfoWeek, out of only three pubished. NewsWire contributor Andreas
Schmidt wrote under "Forever Young" that "For me, it's
not surprising that the HP 3000 goes Internet. In my company, and HP
3000 has served part of our intranet for at least three years. So
what's old about the HP 3000? The system remains young -- that's why
it has lived for more than 25 years."
3000
solutions provider Birket Foster had his letter published as well
under the title "Old Reliable." He said, tongue firmly in
cheek, that "I'm surprised Hewlett-Packard let you write about
the best-kept secret in corporate America." Foster added that
"The fact that it's so reliable, and users measure mean time
between failures in years, makes it invisible to management of
companies that use it."
Wide editing tool available for free
Wyell
Grunwald of Alcoa Engineered Products is offering a free utility to
split large record-width e3000 files for easier editing. A teaser
post on the Internet on Friday reported, "WIDED allows you to
split apart a large record width file, KSAM file, SPOOL file, or
TurboImage dataset into smaller record width file pieces. After
editing these small record width file pieces, WIDED can then be used
to combine the small record width file pieces back into a large
record width file (less than or equal to 9999 characters wide) or
into a TurboImage dataset. WIDED can be used to change key fields in
a dataset, or unload / reload a database to disk files. You can get
your free copy by e-mailing him at Wyell.Grunwald@Alcoa.com.
Smith-Gardner gets Dream-y with WebOrder,
integrates WAP
Online retail application provider Smith-Gardner announced a
pair of improvements to its WebOrder app that's selling new HP
e3000s. The software will be integrated with the Dreamweaver 3 Web
page creation suite from Macromedia. S-G signed the OEM deal to
integrate a visal editing solution that complements the front-end
publisher of WebOrder. The Enterprise Edition of WebOrder has a
WebOrder Publisher component, HTML template files and CGI scripts
designed to deliver an out-of the box e-commerce storefront. Now S-G
will include a WebOrder pallette for Dreamweaver, so WebOrder sites
"can easily drag and drop dynamic data from WebOrder onto their
Web pages." The application uses a WebOrder gateway (NT system)
and the Publisher to communicate with the HP e3000, which S-G calls
the WebOrder Commerce Engine.
Smith-Gardner also announced it was supporting the Wireless
Application Protocol for WebOrder, so wireless devices such as cell
phones and PDAs can communicate orders to WebOrder sites. The
technology promises to let retailers communicate real-time order and
inventory status to customers, as well as shipment confirmation. One
application of WAP in WebOrder might let a company automatically
inform customers it's time to replentish products. McIntyre and King,
a home shopping fulfillment company in the UK, expects to be one of
the first S-G customers to take advantage of WAP.
The
application provider also announced a deal to supply Urban
Outfitters, a retailer with more than 50 stores and the Anthropologie
Web site, with its WebOrder solution.
Hidden Value: getting $STDLISTs
We
got a note from a reader concerning our Hidden Value item in the
February issue about getting the names of $STDLISTS via programs.
Brinton Butler of The Wellness Plan in Detroit wrote, "about
getting the $STDLIST for a known job, the SHOWOUT command will give
the spoolfile ID if you use the format: SHOWOUT JOB=#J1234. I'm not
sure if this can be done programmatically, but I believe it
can."
What's holding up traceroute? Token Ring
Traceroute, a great utility found in network toolboxes on
many Unix systems, has been ported to the HP e3000, but it hasn't
been cleared for takeoff in the customer base. The utility went into
beta test in July of last year. In an online chat event on
3kworld.com this week, customers reported that the tool is being held
up because the current version aborts e3000s which are using Token
Ring. HP pulled the patch containing traceroute until it can clear up
the system aborts it's causing.
Guidelines for building Web applications on
e3000s
Duane
Percox, one of the founders of e3000 application provider
Quintessential School Systems, has offered a great guideline for how
to build applications which interact with Web servers. It's a subject
that Percox knows a lot about, since he's written a Web server for
the HP e3000 (QWEBS) as well as integrated the QSS apps with Web
interfaces for K-12 school districts. Here's the Percox plan for
getting the most out of integration with the Web:
I saw
[an Internet posting] regarding connecting a classic MPE/TurboImage
application to Apache. This information was supplied by Mark Bixby,
who has been instrumental in getting Apache ported to MPE/iX:
Mark
said, "In the shell, you can do this: callci 'file
mydatadb=mydatadb.group.acct' ./mypgm But the problem with this
approach in an Apache CGI environment is that :FILE commands have
job/session scope, i.e. all Web requests currently executing will be
sharing the same pool of :FILE commands because Apache is just a
single job with multiple children. Unless you're very careful and
have a controlled CGI application mix, you can get into trouble doing
this when one CGI decides to do: file
mydatadb=mydatadb.somewhere.else. It's far safer to have your CGI
programs open fully qualified file/database names that don't rely on
file equations."
My
comments (very soapbox-like): This is NOT a specific Apache or Posix
problem, but a problem that ANY web server has operating on MPE/iX.
The "other" Web server (QWEBS), which runs in the MPE space
(not Posix), has this same issue. This is because of the way MPE
treats job/session space as global for file equations and MPE
variables regardless of the number of processes executing in that
job/session.
This
leads me to point out that trying to do CGI directly connected to a
Web server is NOT the best way to produce large/scalable/robust Web
applications. Consider the following issues:
1.
Only one web server can be running on the HPe3k handling standard
port-80 web traffic at a time. This is a fundamental design of
network (tcp) programming and not a mpe/ix limitation.
2.
The one web server running on your HP e3000 will most likely NOT be
situated in the account where your application is resident.
3.
You most likely have more than one application, crossing many
accounts, that will need to interface with a web server.
4.
You will have many security issues opening up access across all these
accounts to give your web cgi access to your existing production
databases.
5. If
you build your cgi tightly coupled with a web server what happens
when you want to support (or need to) another web server?
6.
Classic CGI with its process creation model is very system resource
intensive, and can quickly overwhelm a system when confronted with
lots of simultaneous hits.
7. In
the Unix/NT world it is NOT normal for application developers to have
the ability to do CGI directly with the Web server. This is tightly
controlled for security/performance and access reasons. While the web
servers allow it, the Web administrators restrict this access.
With
these issues raised, you are highly encouraged to build your
production HP e3000 resident Web applications around the following
basic guidelines:
**
Host your web applications in the normal application account.
**
The CGI you write should be thin "glue" that links the Web
transaction from the Web server to your Web application. This is best
done using message files or other forms of IPC (inter process
communication) available to developers on mpe/ix. If your Web server
supports it, develop your "glue" as a callable routine that
is accessible by the Web server without the need to create another
process.
**
Isolate the particulars of the transaction source (Web server in this
example) to the thin "glue." Keep all of this specific
knowledge as far away from your application code as possible.
**
Think of your application as an "application server," which
is waiting for the next transaction. Your thin "glue" CGI
passes along the Web transaction and waits for the results, which are
passed along to the browser. If architected well, this application
server could also receive transactions from other sources beside a
Web server. For each source, you would develop the thin
"glue" which connects the transaction source with your
application server.
**
Design your "application server" to have multiple instances
(multiple processes) of the server code to be able to handle lots of
simultaneous hits at one time.
** If
you have any spare time, consider developing a way for the Web
transaction to be sent to a completely different HP e3000 system
which is hosting your "application server." This would
allow you to distribute Web application loads across multiple
systems.