All of us at one time or another have an obsession with speed. With me, it naturally comes into play with network throughput: I know, an odd hobby but it comes with the territory. And many of you share my interest as well and would like your networks to perform faster. Over the years we have come up with a variety of ways to increase network throughput but certainly the simplest way is to prevent any users from having network access. That isn't very practical or very politically acceptable these days. So, second best is to examine carefully the bottlenecks and fix them. That requires skill, some good tools such as packet analyzers, and a lot of time -- a hard combination in these days of downsizing support staffs that are already stretched pretty thin. Most of us though take a third route, beefing up places where we feel the bottlenecks already exist. While more satisfying than crawling around one's wiring closets with a Sniffer, it can be costly, time-consuming and career-limiting -- especially if your gut feelings on your network bottlenecks turn out to be wrong. One arena that has received a great deal of attention has been upgrading your network cabling to handle faster, 100 megabit technologies. And this year many products have become widely available from a variety of vendors, always a sign that a market is maturing. There has been a great deal written on 100 megabit networking, and I won't give you a primer here: suffice it to say that there are two competing technologies: 100 Base-T, also called Fast Ethernet, and 100 VG AnyLAN, which handles both Ethernet and Token Ring topologies (although my fellow columnist Metcalfe would take issue with using Ethernet with 100VG in the same sentence, most everyone else lumps them together). In these next three columns, I look at 100VG technology. A future series will examine my efforts at using Fast Ethernet. My site for the 100VG network was perhaps the most unusual organization on my travels to date: the Campus Crusade for Christ world headquarters in Orlando, Florida. Campus Crusade is an international missionary organization that has about 600 employees in Orlando and thousands more scattered around the globe. Their data processing needs are immense to support a wide variety of publishing, conferences, fund raising, and other charitable works. Their offices are located in a sprawling one-floor office building that has wiring closets located in different quadrants of the building, with each closet connected back to the central data center in the middle of the building via fiber and several twisted pairs of copper. This is a good thing, because it allows all sorts of flexibility when it comes time to configure the network layout. And flexible is one word that applies to the Campus Crusade network. They are the ultimate equal opportunity topology employer and have tried just about every kind of wire known to grace a network. Before I arrived with my 100VG equipment, they were using16-megabit token ring along with Arcnet and 10Base-T Ethernet and even some Thomas Conrad TCNS 100 megabit wiring thrown in for good measure. There were all very good reasons for choosing one topology or another for the organization. "Well, we started with Arcnet when our offices were located in southern California," said Tim Taylor, manager of network consulting for the organization. When the organization first got into LANs, they were still a mainframe shop with lots of coax in the walls: Arcnet made the most sense since it uses the same RG62 cabling that a 3270 tube has. "When we moved to Orlando four years ago we began to install token ring networks," he told me on my visit there this past June. Campus Crusade runs its operations on three different types of servers: NetWare 3.11 and 3.12, which they will later this year upgrade to 4.1; Santa Cruz Operation's Unix Open Server; and an IBM 4381 mainframe which they are in the process of getting rid of next year. All of their servers, except a NetWare for SAA mainframe gateway, have two network adapters in them: a token ring card, which connects to a backbone ring; and several TCNS cards, which connect to various wiring closet servers running Novell's Multi-Protocol Router software. They use these wiring closet servers to bridge between all the various topologies that are installed in that quadrant of the building, making it easier to transition from one topology to another. They have two servers each running SCO and NetWare: one of the SCO boxes is a Tricord ES/5000 with 192 megabytes of RAM and 8 gigabytes of disk. This beast runs their Oracle database server and is being loaded up with more and more data that was originally kept on the IBM mainframe. Finally, most of their desktop machines are Intel machines running either DOS or Windows, and they do have a few Macintoshes as well. Next week we'll talk about the conversion to the 100VG network. 30 If you are planning about upgrading your NetWare network to 100 megabit techologies, my advice is have an CNE on hand who really understands multi-protocol networks and knows his or her way around an AUTOEXEC.NCF file. Without this expertise on-site, you are in for a rough ride. I was working with such a fellow at the world headquarters of Campus Crusade for Christ in Orlando, Florida. As I mentioned last week, Campus Crusade has one of every kind of wire scheme in their walls: Ethernet, Arcnet, and Token Ring. The wunderkind was Kent Keller, who had designed their network to handle a graceful transition from older, slower technologies to the brave new world of 100VG AnyLAN. Keller, along with the rest of the IT staff, realized that their 16 megabit token ring backbone was getting bogged down. Users were running more applications on a Tricord SCO machine, using PeopleSoft's applications and other Oracle databases. They began an investigation late last year into faster networking technologies. "We settled on HP's 100VG AnyLAN because we don't have category 5 wiring in our walls and because some of our wiring closets are located more than 100 meters apart," said Jerry Hertzler, the network services manager at Campus Crusade. Both of those reasons prevented them from choosing 100Base-T or fast Ethernet technology. So our job was to obtain the necessary parts to put both NetWare and SCO servers on 100VG networks, along with several workstations that required faster throughput for their database queries. At the time we started this project in early January, only Hewlett Packard was shipping 100VG equipment. (Now Thomas Conrad has joined in, along with a few other vendors). We needed several hubs, along with an assortment of ISA, EISA, and PCI network cards to upgrade their network. Campus Crusade is buying PCI-based workstations now, although they had quite a few older ISA machines. And all of their servers used EISA adapters. Keller's design was to install several 100VG NICs in each server in place of the Thomas Conrad adapters to minimize hops from the desktop to each server. Each NIC connects a server to a particular wiring closet, where a NetWare Multiprotocol router is used to connect all the various topologies installed on the desktops in that part of their sprawling building. Keller's idea was to keep the existing wire plant useful as long as possible, while at the same time allowing the more demanding users to receive 100 megabit throughput from end to end. That was the theory, anyway. Our installation wasn't without its share of problems, of course, and that's where having Keller, along with others with complementary experience, helped a great deal. Our first problem was getting the network adapters to work properly. We began by trying to install the PCI cards in a test workstation, to make sure that everything worked. They didn't. Turns out there is a conflict between EMM386 and the PCI drivers. Using a newer version of EMM386 from Microsoft fixed this problem. Next issue were the EISA adapters. HP has done a sloppy job with these: you have to configure them three times: once as part of the base EISA configuration utility when you first put the cards in the machine. Next, you boot the machine in DOS and run HP's configuration utility, making sure that the interrupts and memory parameters that you specified in the EISA configuration are the same in the HP-configuration. This is an unneeded step if the HP NICs were true EISA cards: they should be able to set themselves accordingly and without the need for a separate DOS setup routine. Finally, you have to set the very same parameters in the network operating system (NetWare does this in the INSTALL utility, SCO does this in their netconfig utility.) Of course, given this complexity, you would think mistakes are inevitable, and we managed to mess up on one of the cards. It took a few hours to track down, but because the mistake was found during the work day, he had to wait till off-hours to bring down the server and reconfigure the machine. Another problem was the Macintosh support. After lots of consultation with Novell and HP support engineers, we discovered a bug in the drivers dealing with AppleTalk support on NetWare 3.11. Once we got new drivers from HP, we were off to the races. We had another slight problem, though: some of the workstations were using a moldy version of Novell's shell, NET3.COM. This wouldn't work with the 100VG configuration, and we had to upgrade these machines to the more modern NETX.EXE shells. All this was well and good, and for the most part the Campus Crusade staff got their NetWare servers up and running on 100VG with a minimum of fuss and bother. Next week, I'll talk about the SCO installation and our results. 31 The most interesting thing about testing 100VG AnyLAN networks at the world headquarters of Campus Crusade for Christ is that we ended up testing the wrong things but the users got a faster network anyway. It is nice to have a happy ending to our saga, but first we had to get through some power problems. "Orlando [where the organization's headquarters is located] is the lightning capital of the world," said Tim Taylor, the manager of network consulting for the organization. "We get power outages all the time here, and that's why we have such a huge UPS in our data center." However, the UPS didn't help them when it came time to upgrade their Tricord SCO server to 100VG. Campus Crusade runs on two SCO boxes: the other one is a 486 clone. They tackled this one first and found all sorts of problems with the box's EISA configuration utility. One trick they discovered was to load SMARTDRV to cache the floppy disk: that cut down the installation from over a hour to mere minutes. On to the Tricord. We began at 3 pm one afternoon. When we tried to power the machine back up (after turning it off and installing four 100VG cards in it), it wouldn't start up. Tracking down the problem took the better part of an hour, and luckily we had lots of help. Tricord has fairly strict maintenance agreements, and in order to service the machine you have to have someone that has graduated from their program. None of the Campus Crusade staff had these credentials, but their local Tricord rep, Michael Brown, did. Tricord has some interesting redundant components, and this particular model has four duplicate power supplies. All four were working just fine. One place that the machine doesn't have any redundancy is the place where the power cord enters the machine: after all, you only have one place for it to come in. Unlike other PCs with the standard three-prong adapters, the Tricord uses 220 volt power and a different cord. It took some time to figure out that the cord was working fine. Brown ended up taking apart the power control modules in the machine, and determined that one of the two modules was broken. He decided to replace both of them, rather than spend any more time tracking down the problem. Fortunately for us Tricord maintains a parts depot nearby in Orlando, so it was just a matter of a few hours' wait until these parts showed up. By 8 pm we were jamming. Once the machine was powered on, we configured the adapters and got everything connected. We did some tests to make sure the server could be seen at various places on the network, and that was working fine. Now to test for throughput: was the 100VG network any faster than 16 megabit token ring? We did a simple ftp file transfer from a NetWare server to the SCO box, and found five megabytes of data took about as long with a 100VG connection as it did with a token ring connection. Either our tests were testing the wrong thing or else something was wrong with our network. Since everything else appeared to be working, we decided to call it a night and head home. The next morning we arrived to users with smiling faces. Database queries took about a third less time with 100VG connections between the client and server than they did under token ring. It looked like we were testing the wrong thing, and the users' own experience showed us that we needed to redesign our tests. When later that week we did our own tests of database queries (with everyone off the network), we found a query that took 22 seconds on an end-to-end token ring connection took only 15 seconds on an end-to-end 100VG connection: clearly, the 100VG is faster. And the query interestingly enough, took the same time whether an ISA or PCI 100VG adapter was used: showing that the PCI driver isn't yet up to snuff, we were testing the server side more than the client side, or else HP can write very good ISA drivers. With all this work, do I feel that 100VG is ready for prime time? I would say yes, with some caveats. First off, you should make sure that you have the latest drivers from HP. I also want to put a plea here for HP, SCO and Tricord to do a better job working together to make better drivers. Having my attention focussed on this particular combination of products helped to move things along, but the three companies need to coordinate their engineering. Second, be sure you have documented all the bits and pieces of your network, since once you start upgrading the wire plant you'll be surprised at how many places there are to touch. Campus Crusade's staff forgot about a few print servers and diskless workstations that were still running older versions of DOS and NetWare shells -- these didn't work under the new 100VG setup. Third, I still recommend that you purchase a packet analyzer and know your network before you begin: you may find all sorts of bottlenecks that can be resolved without upgrading your topology. And finally, be careful to make sure that you are testing the right things: as we found out, file transfers don't really do the job. Products Mentioned: AdvanceStack 100VG Hub-15 J2410A 15-port hub $3199 10/100VG Selectable PC LAN Adapters: J2573A (ISA), $229 J2577A (EISA), $279 J2585A (PCI) $249 Hewlett Packard Co. Santa Clara, CA 95051 800 333 1333 800 333 1917 fax http://www.hp.com (web), info@hp.com (email)
Click here to return to the previous
page
David Strom
Port Washington, NY 11050 USA
US TEL: 1 (516) 944-3407