Volume 14, Number 11 - November 1998
by Ron Roberts
Twenty-three members attended the October meeting. Jack McKinney introduced the officers and spoke about the new CACTUS Intel-based system. It's a dual Pentium II, 300 megahertz, with 256 megabytes of memory. He wondered what we're going to do with all that memory. Lindsay jokingly suggested that we could run NT. It has an ultra-wide/fast SCSI bus with two 9 gigabyte drives. The network card is a 3Com 3C509B PCI 10/100. Jack tried to get some consensus on an installation party, but no definite date was nailed down.
The evening's speaker was Clay Boyd who's been working with the Distributed Computing Environment (DCE) for ages. DCE is now an open system in the public domain. It came from the Open Foundation (formerly known as OSF). DCE is an API to support the development of client-server systems. Prior to DCE, everybody was rolling their own using RPC, and they were all mutually incompatible. It was a corallary of the NIH (not invented here) syndrome. IBM invested 450 million dollars in development of DCE, so AIX was chosen as the reference platform.
One of the problems that DCE addresses is the choice of servers. If you have two, which do you use? It uses the portmapper for dynamic identification of IP and port numbers.
Another issue for client/server programming is data typing -- how to reconcile differences between floating point representations, word size, big-endian little-endian, etc. Sun invented the RPCgen compiler to create NDR (network data representation) and XDR (transfer data representation) to handle these differences. Actually, the first RPC was invented by IBM -- called APC for mainframes. These translation routines are not very smart, however. Apollo used NCS -- the compiler that DCE uses.
Bob defined a server as a collection of routines. Using some koosh balls, he proceeded to illustrate some typical routines: put, take, drop. DCE uses a UUID (universally unique identifier) to verify compatibility between client and server. It generates stubs that bundle the procedures on the wire. The IDL compiler generates these stubs.
There are shrink wrap applications such as ENCINA. CICS is built on top of ENCINA. CDE 1.2.2 includes the Distributed Filesystem (DFS), which grew out of the Andrew File System (AFS). It's available for a mere $25 per copy for commercial use. The client identifies the server by using the Cell Directory Service (CDS). The CDS daemon hands YP and White Pages.
One of the goals in client-server systems is to require zero client changes for server scalability. IBM has 9,000 client machines on DFS with 60,000 users and only 5 1/2 administrators. The design point for NFS was ten machines. Microsoft believes that a large enterprise system is fifty machines.
Thanks to Clay Boyd (clay [at] lworks <dot> com) for an extremely informative and entertaining presentation.
by Lindsay Haisley
We're making excellent progress on our new online system for cactus.org. Jack McKinney got a very good deal on a Intel Dual Pentium II box with two 9G hard drives. Our install party was sparsely attended, but Jack, Gil Kloepfer and I assembled at TRMG and installed Linux Slackware 3.5 on the system. After the initial installation we adjourned to Jack's house and spent a good part of the evening trying to get the system's 3C905B network adaptor card working with the Linux Vortex driver, with little success. The new Slackware Linux also presented us with some problems providing a development environment, parts of which are apparently still in beta development.
The new box came home with me to FMP Computer Services headquarters and had Slackware Linux replaced with GNU/Debian Linux v2.0. The network card and the development environment are both now working properly. We've recompiled the kernel to use SMP to take advantage of both processors and have mirrored one 9G hard drive onto the other using RAID-1 (mirroring). The system is now running smoothly on a temporary IP address and will be moved to it's home at Outernet Connection Strategies as soon as possible.
We have nearly 5G allocated for user accounts in /home on the new system, and as soon as possible we'll publicize the new address and start setting up user accounts on the system for all current CACTUS members. To stay informed on our progress, make sure that you're subscribed to the cactus- news mailing list, which you can do (if you haven't already done so) by sending a blank email to cactus-news- subscribe [at] cactus <dot> org.
The new system is quite fast and can compile its 400K kernel in a little over two minutes. Its speed is substantially better that that of our Sparc II. I ran a couple of simple benchmarks to compare the two systems and the results are dramatic!
Sparc II Intel Pentium II --------------------------------------------------------- Gzip of 1.2M 13 seconds 1.3 seconds tarred source Compilation of 4 min, 54 sec 23 seconds same source
This amounts to about a tenfold increase in speed - an entire order of magnitude!
We will be putting all CACTUS member database operations on the new system and plan to start work soon on a new comprehensive database system to handle memberships, newsletter mailouts and online account creation. This will be a substantial programming project involving work in perl, HTML, PHP3 and SQL. We hope to set up a working committee from the membership to help with this project. If you have programming experience or abilities and would be interested in contributing, please send email to fmouse [at] cactus <dot> org.
To renew your membership, please send check or money order payable to CACTUS ($25/yr for regular membership and $96/yr for corporate sponsorship):
PO BOX 9786
AUSTIN, TX 78766-9786
You can also pay in person at the general meetings. The date on the upper right hand of the newsletter mailing label indicates when your membership expires. Please direct any inquiries or address changes to membership [at] cactus <dot> org.
CACTUS meets on the third Thursday of each month at the UT Applied Research Labs (ARL) in the JJ (Jake) Pickle Research Campus (JJ PRC). We'll meet in the main auditorium located directly behind the guard's desk and main lobby.
Please do not show up earlier than 6:20 pm on the specified day. Enter through the main entrance at 10000 Burnet Road for ARL:UT. Tell the guard that you are here for the CACTUS meeting. You will be required to sign a log book, but not required to wear a badge. The guards will direct you to the auditorium entrance. Limited parking in the front of the building is available, but more extensive parking is available in the large parking lot just north of the ARL building. After 6:30 pm, all entrances to JJ PRC, except for the Burnet Road entrance, are closed and locked. You can still enter the parking lot in front of the ARL building. No parking tags are necessary after 6:00 pm. See maps for further details.
Online maps are available at:
As always, please leave the facility as you saw it when you arrived. So let's clean up all pizza and soda items before we leave. Thanks and hope to see all of you there.
| | ^ <---- to MOPAC | | | | | | North | | B | to Braker Lane ---------------+ | U -+ /-----------+ | R | | | | N | +--------+ | | E | Parking | | | T | Lot | | | +----------+ | | R | | D +------------+ | | | | +-------+ | | | | +---+ | | | | | | | Rutland | ARL | | | | +--------- | | | | | +--------- | | | +---+ | | | +---+---+ | +------------+ | | | | | | South | | | to US 183 | | v