rlk@tusun2.knet.utulsa.edu ( Richard L. Kruse II) (06/24/89)
After sifting through the requests for more info on this subject, and receiving some pointers on summary posting (thanks, Thomas Brisco!), I decided to make a stab at posting a summary of the replies I received (I actually received 26). Thanks again to all of you who took time to respond. The replies helped us make an informed decision. The decision was basically between MultiNet from TGV and WIN/TCP from The Wollongong Group. What was the decision, you ask? Well, we finally chose the MultiNet product from TGV (we are actually dealing with SRI). Some of the factors used in making the decision included availability of source code (thus making patches over the net possible and easy), UNIX-like utilities (RPC, FINGER, WHOIS, LPD, R-commands, TALK), and the pricing. Let me point out that both Wollongong and TGV offer NFS servers, GATED routing and SNMP, as well as the standard TCP/IP stuff. I was pleased and impressed that representatives from both of the top contendors, TWG and TGV, took time to reply to my query as well; nice to see them involved! To assist any of you who might need this, here are the contacts I know of for the "top two": TGV: Desiree@Warbucks.AI.SRI.COM (Desiree Champagne) TWG: djw@twg.com (Dennis Wong) [I hope this is the right person - we actually dealt mostly with DEC re: Wollongong - rlk] Rick Kruse University of Tulsa Now for the (long) summary (drum roll here...) ****************************************************************************** As I stated previously, the top vendors of choice here seem to be TGV (with MultiNet) and Wollongong (with WIN/TCP). There were a few votes for others, including Fusion, Ultrix and 4.3 BSD (I *wish* we could run U**x ;-}). ****************************************************************************** From: enger@sccgate.scc.com ( Robert M. Enger) We use wollongong. [Some stuff deleted here. -rlk ] The Multinet product (that used to be available from SRI) is supposed to be good. Kashtan (?) and another person are supposed to have left SRI to continue development of Multinet for commercial sale. Maybe SRI can direct you to them for further info. The other little guys are in general WAY WAY OUT OF DATE. They lag way behind the latest internet engineering. if you want discriminators, ask possible vendors how much of the Jacobson/Karels/Karn upgrades they have implemented. slugs like Bridge and CMC won't even know what you're talking about. While wollongong hasn't done the header template prediction stuff, they have done the slow start/congestion control stuff, a good sign in my opinion. ****************************************************************************** From: "Louis A. Mamakos" <uunet!trantor.umd.edu!louie> Run Ultrix on your VAX 6320. It has all of the TCP/IP networking you need for free already included. ****************************************************************************** From: uunet!proteon.com!jas (John A. Shriver) I think Wollongong remains the best. Forget Network Research Fusion, it's been a cruel hoax in my experience. You might want to look at Process Software's package, its doesn't have all the bells and whistles that Wollongong does, but it's solid. Also, they are small enough that support comes from the developers, which is a big win for you. ****************************************************************************** From: Ed Frankenberry <uunet!BBN.COM!ezf> Have you talked to DEC about VUX "the VAX-Ultrix connection"? It was in beta test the last time I checked, and may be available now. It is/will be a standard layered software option. [This wasn't available when we talked to DEC, and also didn't provide the flexability we were looking for - seemed to use a Ultrix machine as a gateway, and from our conversations the product seemed to add to the complexity of networking, just what we *don't* need. -rlk] ****************************************************************************** From: Yehavi Bourvine +972-2-584279 <uunet!CUNYVM.CUNY.EDU!YEHAVI%HUJIVMS.BITNET> I have exactly the same problem as you have. We are buying a VAX-6330 and have to support more than 100 incoming TELNET sessions. The replies I got for my message specify Wollgong and MultiNet as the ones that will do the job. We've finally choosed MultiNet (I've ordered an evaluation copy to test it before we buy it) because it seems (according to the publications) better and comes with full sources (and has site licence in good price...). I enclose here all the replies I got, including my correspondence with the MultiNet people. [Interesting replies included... -rlk] From: pack@acdpyr.UCAR.EDU (Dan Packman) We run Wollongong 5.0.1 (the first SMP version of their product) and if anything it is more robust than previous versions under VMS V4.x. Their early products were seriously flawed, but I can wholeheartedly recommend the current release. Dec has muddied the waters by announcing their own tcp/ip implementation that initially will only allow Decwindows to operate but might eventually have a user level (ftp, telnet, etc). Dec's product might ultimately prove the best and most seamless, but right now they don't have a complete product. From: Lori Corrin <purple@hadrian.uwo.ca> I'm the person around here responsible for our multinet software. We tried running the fusion software around December 88 and there were still SMP problems. Multinet seems to have solved them. We did have to run through a few patches for some wierdness but it never took more than a day for me to get a patch ftped to me. We tend to use rlogin from our terminal servers (we have a number of unix boxes). The rcommands are part of the distribution. Along with some other real nice features. From what we could tell the system seemed to slow down when we hit 110+ users (as a 6230) but we were never able to determine if the slowdown was the cpu or multinet although I'm tending to suspect the cpu since I haven't seen the problem since we upgraded to a 6330. I would expect approx the same performance from telnet since they are both implementing through the multinet kernel. The annoyances tend to relate to signal processing (ie ^C doesn't respond immediately but after the next buffer). We have to put a monitor on it to find that problem. The other problems involve some wierdness to our suns in ftp. The one thing lacking in the Multinet I find is the documentation they shipped me an improved version without the improved documentation but I knew that when I asked for it. The existing documentation works fine but some of the new features aren't documentated. (like rlogin) My copy included source and I used that if I had problems following what was going on. Multinet also has an NFS server side available and are working on a client side I believe. From: Lori Corrin <purple@hadrian.uwo.ca> We have some problems with flow control. With ^O/^C the abort output flag doesn't seem to be a priority packet it waits till the current buffer is done. This could be a problem with our terminal server, we're going to put a monitor o n it to see who's not doing what. Also ftp doesn't support type tenex which is a bit inconvient. and we've found a wierd problem that happens to our suns were a ftp connection is made to the suns and every other user command (in the same ftp session) takes a few minutes to respond with the password prompt. Again we're not sure were the problem originates with this one. With the remote printer service an ^L seems to get sent at the beginning of a job to the remote printer which was screwing up our postscript. We're told that one is a VMS symbiont problem so wrote a simple filter to check the first couple of characters and see if it's a postscript file and if it is strip out the preceding characters. The sources are in gnu C (they supply gnu C as well.) but it seems to use some of the VMS libraries. It supports a variety of devices: ethernet, arpanet, x.25 and chaosnet. It also handles ip over decnet and shared vms ethernet. It will run decnet over ip as w ell. Those are the ones in my manuals If you're looking for something specific you might talk to ken adelman (adelman@tgv.com) he's the technical support for m ultinet and they may have some devices available that aren't in my documentation From: adelman@TGV.COM (Kenneth Adelman) > 1. I guess the MultiNet can use a dedicated ethernet controller (DEUNA etc). > without running DECnet on it. Right? Correct. And the procedure is exactly the same as if DECnet was running. > 2. Is it possible to remove and re-use (as the system's manager will) the > various utilities, like some of Rxxx, or replace the SMTP server with > my own one? Yes. You can replace clients by defining symbions to invoke yours, and replace servers by using the server configuration manager utility to specify a different image to be run. > 3. I heard that the sources are a standard part of the package. Since it was > not mentioned in the SPD, is it still the case? The SPD I gave you is for the product purchased from TGV. You should purchase from SRI International, as they have a better educational price and include sources at no additional charge. >From: David L. Kashtan <KASHTAN@SRI-IU.ARPA> > >1) What hardware is needed to connect a Vax VMS system to a Tcp/Ip network? > That depends on what the communication medium is going to be. > Assuming that this will be Ethernet (that is the most common medium), > MultiNet will work with any of the following hardware interfaces: > a) DEC ethernet (DEUNA/DEQUNA/DELUA ...etc). MultiNet also > can share this ethernet interface with other networking > protocols (.e.g. DECNET) > > b) Interlan NI1010 ethernet interface. > > c) 3-Com ethernet interface. > > d) Excelan EXOS ethernet interface. > > If what you need is something OTHER than ethernet, there are a wide > variety of communications controllers that are supported by MultiNet. > Just let me know what you are going to be using as the communication > medium. > >2) How much does it cost? > You will have to get the pricess from the various > manufacturers. Most VAXes nowadays come with an > ethernet interface -- so there is no extra hardware > cost for MultiNet. If you need to buy an ethernet > interface the following may be of interest: > DEC -- theirs tends to be the most expensive > but support should be good. > > Interlan -- Less expensive than DEC. We have > had good reliability with these. > > 3-Com -- No longer manufactures these -- you would > have to get them on the 2nd hand market. > Not recommended. > > Excelan -- Their interface has On-Board TCP, which > MultiNet cannot make use of (it runs its > own TCP in the VMS kernel). Thus, you are > probably paying for things that you don't > need. > >3) What software is required to be in VMS in order for MULTINET to run? > None -- it runs on a "vanilla" VMS system. > > To recompile any of the user-level code will require the DEC > VAX-11 "C" compiler. It is unlikely that you will need to > recompile anything. > > To recompile the networking kernel will require either a VMS system > that also has Eunice (A UNIX system layered on top of VMS) or a VAX > running 4.2bsd or 4.3bsd UNIX. It is VERY unlikely that you will > need to recompile the networking kernel. > From: "RANDY CATOE" <randy@TWG.COM> Subject: RE: TcpIp software for SMP VAXes and large number of incoming Telnets. To: "yehavi" <yehavi%hujivms.bitnet@cunyvm.cuny.edu> Wollongong routinely tests new software at a commercial database development company that runs a cluster of six or more 6360 VAX systems. The normal load on those systems is in excess of 300 telnet users each. The TCP/IP software is WIN/TCP. The WIN/TCP networking kernel differs from its competitors in that we have built an SMP implementation that accomplishes multi-processor synchronization on a per connection basis. The result is that we incur far less multi-processor synchronization overhead than in other approaches. This results in more processor resources for other computing. From: "IVAX::IJAH400" <ijah400%ivax.decnet@gold.bacs.indiana.edu> I wasn't directly involved in the evaluation, but we are running the Wollongong WIN/TCP package on the recommendation of the systems group at the main Indiana University campus in Bloomington, IN. We haven't had any problems with it yet, but most of our terminal connections are still via LAT (here at I.U.P.U.I. in Indianapolis). Also, our load is rather light compared to BACS in Bloomington (50-60 users on an 8800). We are still running VMS V4.7 (to go to V5.1 this summer). We are running version V3.2 of the WIN/TCP product. BACS at I.U. Bloomington is using Telnet rather heavily in a cluster of 2 8820s running VMS A5.0-2 and WIN/TCP V5.0, and one each of a 11/780, 11/785, and an 8650 running VMS V4.7 and WIN/TCP V3.0 (I think, maybe V3.2). Their loads range up to about 100 users on each of the 8820s, 40-50 on the 11/78xs and 80 or so on the 8650, with more than 90% of the connections by Telnet. BACS did some pretty thorough testing on each of the three products you mentioned and Telnet performance was a very important issue. Also, whatever they decided on was going to be recommended to other sites all over the indiana-net. There was also a side issue about which vendors could get their software working the smoothest under 5.0 on the SMP machines (the testing was going on early last fall). I don't know how cooperative the TCP/IP vendors are over there, but BACS was testing this stuff for a couple of months, and I think they got some kind of temporary evaluation license deal from the vendors being evaluated for that period (of course, there were a lot of $$$ at stake :-). Maybe you could get a similar deal; there's certainly no substitute for a hands-on pounding out of the products on one's own system, with one's own load, etc., to make sure you're getting what's best for your site... Attached below is a summary on the BACS evaluation I received from the manager of the systems group there. From: AQUA::FLOWERS "Chuck Flowers" 16-MAY-1989 13:42 In my group, Steve Smail did the testing of the various vendor's TCP products. Fusion was dropped early due to lack of stability. It could not stand up to any kind of load. The battle then boiled down to TWG and MultiNet. Both products have their pros and cons but for a high load system where the maximum speed and stability is needed, TWG was the vendor of choice. The big advantage to MultiNet was the price. In fact (pending Jim Williams completion of the contract) we will be offering MultiNet to our departmental users who need the TCP/IP functionality. MultiNet has offered us a site license which makes it a very attractive package. The TWG package does impact the cpu significantly. In our dual cpu 8820s, we may see 75% (of one cpu) kernel loading with 75 users on the system depending on activity. But it was the best offering at the time of our testing and still is to our knowledge. It is also more expensive than MultiNet but it can stand all the load that we can throw at it! From: "IVAX::IJAH400" <ijah400%ivax.decnet@gold.bacs.indiana.edu> I think by "unstable", Chuck meant that under heavy loads the system would tend to crash. Remember that this was going on last fall when all the TCP/IP vendors were still getting the bugs out of their software for SMP. SMP device-driver synchronization bugs tend to show up (i.e., cause the system to bugcheck and die) more often as the load (processes using the driver or device) increases. The MultiNet product is inexpensive, and the user interfaces are more VMS-native-like than the Wollongong product. However, what one is paying for with the Wollongong product is support. Support for the MultiNet product was limited at the time the evaluation was taking place (sort of like, you report the bug and they'll try to fix it in the next release). Wollongong eventually supplied patches on the spot and got their product to the point where it could stand up under a heavy load, so that's why we're running the Wollongong stuff. Since the product was going to be critical to system availability (since almost all their terminal connects are via Telnet), the stability and support issues were considered to outweigh the issues of how nice the user interface looked and price. Perhaps the situation has changed since last fall. I noticed in the file VENDORS-GUIDE.DOC from the Network Information Center (nic.sri.com) that Excelan Inc. is now licensed to sell MultiNet. From: DESIREE@Warbucks.AI.SRI.COM (Desiree Champagne) In cases of problems you can contact us via email, phone and fax. Fixes can be ftp'd or a new tape will be shipped. (I almost always ship tapes via federal express or courier.) What you heard may be correct in both cases. Depending on what the fixes actually were, we may not have made a general distribution tape and shipped to everybody. It is totally dependent on if is an actual bug that HAS to be fixed or if it is just something that a user wishes us to install... In the case of major bugs that can cause any performance problems, tapes are shipped if necessary; otherwise all users are notified via the mailing list. (For instance, we discovered an SMP bug that affects only the last 25 tapes shipped; we are reshipping to all sites running multiprocessors.) In short, we make every effort to provide service to our clients...and the feedback we have had is that our support is excellent. If you wish, I can give you references that I believe will be compeletely honest about our strong and weak points. From: Peter Marshall <peter@hadrian.uwo.ca> We tried Fusion and ran into problems and we were not happy with the support. We eventually returned the Fusion and purchased Multinet which has had exceptional support (even though we officially don't have any). I'm not sure if we ever got 100 connections before everyone left for Summer, but we were up in the high 80s without great problems. We are currently using "rlogin" rather than telnet for most connections. [Whew! Thanks, Yehavi! -rlk] ****************************************************************************** From: uunet!Kodak.COM!messingr (Rich Messinger x24361 B83, Rm 528, RL, 02221) We have just been going thru an analysis of the same question for our 8800. The other main contender that we saw was the solution form Excelan where much of the work is done on the board. Check them out as well as Wollongong. ****************************************************************************** From: "Paul Pomes (I'm the NRA!)" <drd!texbell!uiucuxc!uxc.cso.uiuc.edu!paul> The best comes free with 4.3 BSD UNIX. [My personal favorite! -rlk] ****************************************************************************** I would strongly encourage you to look into the Multinet tcp-ip package that is distributed by SRI. It was strongly recommended to me by "networking gurus everywhere", and from looking at the flyer describing it and knowing the wollongong product from experience, I think you get more for your money with the MultiNet product. The MultiNet product supports decnet over ip, ip over decnet, slip, lpd and you can even get nfs for it for an additional fee. The wollongong product does not include any of these. Additionally, you can get sources for the MultiNet product, not so with the Wollongong product. ****************************************************************************** From: uunet!THENIC.THE.NET!STUART (L. Stuart Vance) My personal recommendation is MultiNet, sold by SRI International. It is the most fully featured IP for VMS implementation I've run across. I've included the latest release notes for your reference at the end of this message. [I left these out 'cause this is getting long... -rlk] I have several qualms about Wollongong [qualms left out -rlk] You might also check out Fusion from Network Research Corp (NRC). The contact at SRI is Desiree' Champagne (415/859-6083). NRC's number is 805/485-2700. SRI will give you a 30 day trial of the software if you want to check it out. ****************************************************************************** From: uunet!decwrl.dec.com!hwchoy%zpovc.DEC (Life, The Universe and Everything.) I would strongly urge you to look at *real closely* the attached products. I have left the developer's address at the end. Talk to him if you like, he's real nice and technical. ps: understand that I don't represent my company (DEC) in any capacity over this matter. But you're welcomed to quote it as my personal opinion. [Deleted attached MultiNet Spec Sheet -rlk] ****************************************************************************** From: "RANDY CATOE" <uunet!TWG.COM!randy> determining the BEST solution obviously requires criteria. If you choose from a performance criterion, I believe that you will find WIN/TCP from Wollongong a superior choice for SMP machines. WIN/TCP's SMP implementation was developed and continues to be tested with the assistance of test sites where 6360 machines are the norm. In order to minimize multi- processor overhead, WIN/TCP uses a finer degree of granularity for "spin-locks" than other implemenations of TCP/IP (or for that matter DECnet). If you're choice is feature driven, WIN/TCP is continually being enhanced by a team of engineers to include the latest network technology. We have recently upgraded to the lastest BSD compatible versions of network utilities such as BIND and GATED. We offer a VMS SNMP agent for network management, A VAXmail SMTP relay, an ALL-IN-ONE relay, support for IP security options (as well as other security enhancements such as support for VMS security alarms). The Wollongong Group, INC also offers a choice of support agreements, and a full complement of training to select from. ****************************************************************************** From: drd!jarsun1!jensen (peter jensen) I must have missed your original query about TCP/IP on VMS. Anyways, we have a MicroVAX-II running VMS. All our other systems are Sun workstations. Anyways, when I looked into this in the beginning of '88, places like Wollongong and NRC wanted $6000+ for their TCP/IP alone. Since cost was a factor, we decided to go with CMU's TCP/IP. It only costs $150 (still), and has worked out beautifully. It meets all our needs for telnet and ftp. It also has lpr and lpd for remote printing both ways, and mail (which we don't use). It has a $QIO interface to the network routines, which I have used to write little utilities. It also comes with complete source code, but you need a BLISS compiler if you want to make any changes. Quite a few universities use this software already, and I have no problems recommending it. ****************************************************************************** I hope this helps and turns out to be more than just a waste of bandwidth! -rlk