zz1he@sdcc19.ucsd.EDU (Heather Ebey) (10/11/89)
Does anyone know for sure if increasing RAM on a 1MB Plus running Apple's AppleShare File Server and Printer Server software will increase performance? Each server services six workstations and and ImageWriter II. All software is on the server and is shared and accessed simultaneously from six Mac Pluses. We are thinking of increasing the RAM to 2.5 Megs. Does anyone else have a simular lab setup and have experience in performance issues? Is there any software that can be run from a client that will evaluate performance? I will greatly appreciate any input you can give me on this. === hebey@ucsd.edu (also UUCP, BITNET) Voice: (619) 534-2448 Heather Ebey, Microlabs Support UC San Diego, Instructional Computing Center AP&M 1250, C-010, La Jolla, CA. 92093
truesdel@ics.uci.edu (Scott Truesdell) (10/11/89)
zz1he@sdcc19.ucsd.EDU (Heather Ebey) writes: >Does anyone know for sure if increasing RAM on a 1MB Plus running >Apple's AppleShare File Server and Printer Server software will >increase performance? Well, yes... it will help a little bit. >Each server services six workstations and >and ImageWriter II. All software is on the server and is shared >and accessed simultaneously from six Mac Pluses. Ugh! You CAN run AppleShare like that, but that has got to be incredibly agonizingly slow. Plain and simply, LocalTalk does not have 1/10 the bandwidth (well, 1/40 :^) ) to support that kind of operation. It is intended for printer and file sharing. >We are thinking of increasing the RAM to 2.5 Megs. That's a good number, though the performance won't be as much as could be gained by, say, just switching to an SE, let alone a II or SE/30. >Does anyone else have a simular lab setup and have experience in >performance issues? I once had to use a Plus as an emergency server while an SE was getting repaired on one of my low-performance nets. The degradation was very noticeable. >Is there any software that can be run from a client that will >evaluate performance? Well, just using a stopwatch (yes, dear friends, LocalTalk is THAT slow) can yield some good insight into system performance. Heather, I guess you think I sound pretty negative. The extra memory WILL help a little bit and it's a pretty cheap upgrade, but I doubt the clients will be able to feel the difference. Truthfully, the only way it is feasible to run applications off a server without (to most users) agonizing delays, is to go EtherTalk all around. That's pretty expensive. Especially considering printing solutions. Even for straight file service, the Mac Plus has the slowest SCSI port in the whole Mac lineup. The SE has over double the throughput of the Plus through the SCSI port, and the SE/30 and II series doubles that again. Then one would want to hang the fastest HD possible on the server. And still these gains would be miniscule when bottlenecked through the two LocalTalk/AppleShare killers: 1/4 megabit/second throughput and no multitasking. What I mean by no multitasking, is that the server performs all its tasks sequentially, one at a time. It recieves a packet. Stops everything. digests the packet. Stops everything. Processes the command. Stops everything. Goes out to disk. Stops everything. Caches the disk fetches (this is the one place where the added memory would help). Stops everything. Dribbles the cached packets out the LocalTalk line. Stops everything. Acertains correct transfer of packets. Then finally gets around to requesting more data from the server disk. How this translates into performance is: Not only is LocalTalk extremely slow, but 2/3 to 1/3 the time, it's not even being used because it's waiting for the Mac to assemble and digest more packets! I guess there's some good in that scheme: with that miniscule bandwidth, it's the only way other devices would ever have a fighting chance of arbitrating some bandwidth! So. You have got what you have got, as far as networks go, and there is no chance of petitioning for faster servers, let alone EtherNet. What next? Here is a run down of possible alternatives, cheapest first: 1. Adding RAM to the servers is cheapest but the performance increase will be negligible. 2. Adding 2MB of RAM to every client will allow them to run RAMdisk software, transfer programs down to RAMdisk, and run with very acceptable performance LOCALLY. 3. Upgrading the servers to SE/30s or, better yet, old headless Mac IIs will upgrade the performance noticeably over solution #1, but running applications from the server will still be a study in frustration. 4. Adding small hard disks to each client workstation. This would yield benefits similar to solution 2 using more comventional methods at a higher (twice) cost. 5. Ugrade to EtherNet all around. This would make running off the servers tolerable, even pleasant almost, but would cost a small fortune. I hope this helps a little. Scott Truesdell Information and Computer Science U. C. Irvine 714/856-5697 P.S. Of the 200 Macs I have networked in our department, only 5 of them have ethernet cards :^( and there are no AppleShare servers with Ethernet. :^( This is sad. But this will change as part of incremental upgrades I have scheduled through next year. At least they let me have some kick-butt servers! :^) -- Scott Truesdell
scb1@tank.uchicago.edu (Sam Blackman) (10/11/89)
In article <349@sdcc19.ucsd.EDU> zz1he@sdcc19.ucsd.EDU (Heather Ebey) writes: >Does anyone know for sure if increasing RAM on a 1MB Plus running >Apple's AppleShare File Server and Printer Server software will >increase performance? Each server services six workstations and >and ImageWriter II. All software is on the server and is shared >and accessed simultaneously from six Mac Pluses. > I once ran AppleShare and upgraded from a Plus on a HD to an SE with 2.5Mb. The only increase in performance that I noted was a smaller frequency of hang-ups and system crashes. What I tied most of the trouble to was the inability of the hard drive to handle multiple write commands being issued from multiple nodes. Aside from that ... >We are thinking of increasing the RAM to 2.5 Megs. > >Does anyone else have a simular lab setup and have experience in >performance issues? > >Is there any software that can be run from a client that will >evaluate performance? You might want to try InterPoll from Apple or even better, TrafficWatch from Farallon. It does traffic evaluations and includes an Excel macro that will take the data and put it into Excel and graph it out for you (way cool!) >I will greatly appreciate any input you can give me on this. > > >=== hebey@ucsd.edu (also UUCP, BITNET) Voice: (619) 534-2448 >Heather Ebey, Microlabs Support >UC San Diego, Instructional Computing Center >AP&M 1250, C-010, La Jolla, CA. 92093 Hope that helps!! Sam -- Samuel C. Blackman ! InterNet : scb1@tank.uchicago.edu Link : UG0184 University of Chicago ! Disclaimer : Who cares what I say? I'm a student ! 5319 S. Maryland Ave. #2 ! Quote : "Changing the world one person at a time" Chicago, IL 60615 ! Phone : (312) 715-3100 x.60 (w) (312) 947-8652 (h)
laa@beta.lanl.gov (Lee A Ankeny) (10/12/89)
zz1he@sdcc19.ucsd.EDU (Heather Ebey) writes: >Does anyone know for sure if increasing RAM on a 1MB Plus running >Apple's AppleShare File Server and Printer Server software will >increase performance? Scott Truesdell replies: >Well, yes... it will help a little bit. ... (followed by a discussion of factors affecting AppleShare speed.) I'd like to share the results of my own crude test of file transfer speed on LocalTalk versus EtherTalk networks. I timed the transfer of a 2 megabyte file both to and from AppleShare servers attached to each network. Both servers were Mac Plusses, though the hard disks were different. The test was done at night, when network traffic was next to nil on both networks. Timing was done by stopwatch. The results were dissapointing. Transfer rate was between 13 & 15 K-bytes/second, on each network, in each direction. This implies to me that the bottleneck is not in hardware, but in the AppleShare software itself. Maybe the hard disk on the EtherTalk server had a relatively slow SCSI transfer rate, but I doubt that the difference was significant. Based on all this, I wouldn't run out and buy Ethernet just to improve AppleShare. Admittedly this test was less than exhaustive, but I don't think the results are out of line. If anyone out there has done a better comparitive study, I'd like to hear about it. Lee Ankeny laa@lanl.gov
alexis@panix.UUCP (Alexis Rosen) (10/12/89)
In article <8910111605.AA03949@aquarium.ecn.purdue.edu> moyman@AQUARIUM.ECN.PURDUE.EDU (James M Moya) writes: > zz1he@sdcc19.ucsd.EDU (Heather Ebey) writes: > > >Each server services six workstations and > >and ImageWriter II. All software is on the server and is shared > >and accessed simultaneously from six Mac Pluses. > >I think the bandwidth can certainly be a problem, but not with 6 >Mac +'s and a server(also a Mac +). I think scott summed it all up >when he discused the throughput for the different Mac machines being >used as file servers. Correct, in general. It constantly amazes me how people want to spend the right money in all the wrong places... > [...] This fall I added another server (a headless IIcx) and >split the network (physically) in two (20 mac's and a LaserWriter per) [...] Not a great move, generally speaking. It's cheaper and much easier, and *faster*, to use an SE/30 as a server. Especially if you are using Ethernet. I suspect a headless IIci would be faster than an SE/30 but perhaps not by nearly as much as you'd like. That's because throughput for LocalTalk will be limited by the wire and for EtherTalk, by the NIC. Since Nubus is slower than PDS, the IIci's NIC will be slower than an equivalent NIC in an SE/30 (which is why the SE/30 wins over the IIcx, BTW). On the other hand, SCSI on the IIci may run considerably faster than on the SE/30 (and IIcx), and that might make the IIci faster again. I guess I'll know more when my IIci shows up. (I feel like St. Joan- "How long, god, how long?" :-) >With that number of machines(6) I think your server is the major drawback... > >If you want to increase your performance, buy a headless Mac II family >machine (a ci..:-) for a server and a couple copies of timbuktu... No, buy an SE/30 and save yourself the trouble. It's more horsepower than a IIcx and more than any six-Mac network needs. Alexis Rosen {rutgers,harvard,ucbvax,philabs,most-other-places}!cmcl2!panix!alexis
truesdel@ics.uci.edu (Scott Truesdell) (10/12/89)
Lee Ankeny mentions testing LocalTalk and EtherTalk AppleShare implementations and coming up with negligible performance differences. He is not the first person to point this out to me. Someone with experience with EtherTalk on my same campus emailed me to state the the improvement just wasn't that great. Frankly I'm shocked, and I volunteered to come look for bottlenecks. It seems, though, that this lack of blazing speed improvement is not uncommon. The only experience I have had with Macs on EtherNet is on a small Novell network. I was very favorably impressed with the performance of this setup. Very. As in, "almost as fast as a local hard disk". The Novell server is an Everex Step 386 with a large, fast HD (I forget which make). I still think something's fishy with the EtherTalk setups. Unless we are looking at the software being the bottleneck. And, at Ethernet speeds, the performance of a Plus, such as Lee tested, would be quite noticeable. I know good hardware is not the bottleneck. I routinely ftp between my ethernet-equiped Mac and our campus's large UNIX machines at 20k/sec and the same file via [empty] LocalTalk/AppleShare goes at ave. 14k/sec. So what's the difference? Time of day / peak useage. If I were to perform the same transfers during peak useage, the ftp would drop to around 10k-15k/sec while the LocalTalk would drop to it's knees around .25 - .5k/sec or worse. What this shows to me is that the cabling scheme is like the freeway/expressway while the servers and clients and their software are like the on-ramps and off-ramps. Ethernet is basically buying you more lanes on the freeway. I guess this also points out where the ral benefits of EtherNet lie. --scott -- Scott Truesdell
Ravinder.Chandhok@CS.CMU.EDU (Rob Chandhok) (10/12/89)
What was the network configuration ? Was it always between 2 Mac Pluses on the same cable, one localtalk and one Ethernet ? What Ether hardware did you use for the Mac Plus's ? I have two servers, one is a Mac Plus on localtalk, and one is a Mac II on Ethernet. They are bridged by a KFPS-4 running KIP. Here are my experiences with them: * For Mac II clients on the ethernet, the ethertalk server is fast enough to compile off of. Almost as fast as a local disk, I don't have my timings, but when I did them they were an order of magnitude faster than a localtalk connection. (I use MPW and MacApp, and keep a shared copy of MacApp on the server) * A localtalk client will communicate better to a localtalk server (no bridge or gateway involved). * Don't even try and use anything less than a KFPS-4 to bridge between ethertalk clients and localtalk servers, or vice versa. If the bridge drops packets, the retry timer in the AppleShare server is set so long that transfer speeds slow to a crawl. There is a patch to AppleShare to help with this problem, but the real soultion is to get something that can deal with lots of ethernet packets without dropping them. I know, there are other things than fastpaths, I don't have experience with them. * I agree, don't just buy ethertalk for the server, unless you buy it for the clients, you will still be limited by localtalk speeds, which, IMHO, is the real bottleneck. AppleShare over ethertalk is just fine. AppleShare over loaltalk is fine too, but don't get too hyped up about the speed. start with localtalk speed, and add in all the protocol overhead. Maybe Appleshare would be a bit better if it used a stream instead of ATP, but I'll bet localtalk is still the bounding factor. Rob
desnoyer@apple.com (Peter Desnoyers) (10/13/89)
In article <33912@beta.lanl.gov> laa@beta.lanl.gov (Lee A Ankeny) writes: > I'd like to share the results of my own crude test of file transfer speed > on LocalTalk versus EtherTalk networks. > > I timed the transfer of a 2 megabyte file both to and from AppleShare > servers attached to each network. Both servers were Mac Plusses, though the > hard disks were different. What sort of network interface do the servers have? If the server is a Mac Plus, I would think that it has a LocalTalk interface. It doesn't matter how fast you can receive data if the server can only send it at 230kbit/sec. > The results were dissapointing. Transfer rate was between 13 & 15 > K-bytes/second, on each network, in each direction. (approximately 50% of LocalTalk bandwidth. Note - theoretical throughput for LocalTalk is >95% of bus speed.) I just copied a 2Meg file from a server (Mac II, some sort of Northern Telecom big disk with built-in backup) to my Mac II (Apple Ethernet card, using MPW duplicate) and got a data rate of 42890 bytes/sec, or 343Kbit/sec. Not astounding, but better than 15kbyte/sec. Peter Desnoyers Apple ATG (408) 974-4469
demarsee@ICARUS.CNS.SYR.EDU (Darryl E. Marsee) (10/13/89)
>I just copied a 2Meg file from a server (Mac II, some sort of Northern >Telecom big disk with built-in backup) to my Mac II (Apple Ethernet card, >using MPW duplicate) and got a data rate of 42890 bytes/sec, or >343Kbit/sec. Not astounding, but better than 15kbyte/sec. Pete, this proportion pretty much agrees with what we've seen. In timing tests to see how long it takes to launch an application over the network from an AppleShare server, there was an approximately threefold increase in speed when running over EtherTalk than when running over LocalTalk. Darryl Marsee Syracuse University
ajq@mace.cc.purdue.edu (John O'Malley) (10/13/89)
In article <33912@beta.lanl.gov> laa@beta.lanl.gov (Lee A Ankeny) writes: >If anyone out there has done a better comparitive study, I'd like to hear >about it. > >Lee Ankeny laa@lanl.gov MacUser's May 1989 issue has got a great report that compares LocalTalk vs. EtherTalk, AppleShare vs. TOPS, and also covers performance differences when using a Mac Plus, SE, II, IIx, SE/30, and Compaq 286. It's a great report. Required reading for any Mac network manager, I'd say. A short excerpt about RAM on the server: "How much RAM should your file server have? As little as you can get away with. We found little difference in performance between 1 and 5 megabytes of RAM. AppleShare gets along fine with 1 megabyte; however, if you plan to run it concurrently with a spooler or E-mail, we recommend at least 2 megabytes." (MacUser, May '89, p. 160) -John --- John O'Malley / Macintosh / Purdue University / (317) mace.cc.purdue.edu!ajq / Specialist / Computing Center / 494-1787
xdaa374@ut-emx.UUCP (William T. Douglass) (10/13/89)
In article <1989Oct12.114440.27973> truesdel@ics.uci.edu writes: > >Lee Ankeny mentions testing LocalTalk and EtherTalk AppleShare >implementations and coming up with negligible performance differences. We have an Ethernet set-up, based primarily on Mac II & SE/30 computers. The servers are a II & an SE/30, and most workstations are IIs. We have realized a dramatic speed-up in throughput by switching to Ethernet. I consider the cost & set-up effort to be worthwhile. As for the reports of little or no speed-up from LocalTalk to EtherTalk, all I can say is that you CAN realize substantial improvements in file & data transfers by upgrading. File transfers show quadrupled throughput (from 16k/sec to about 65-70 K/sec.) with a II workstation, doubled speed for plain SEs. >I still think something's fishy with the EtherTalk setups. Unless we are >looking at the software being the bottleneck. And, at Ethernet speeds, >the performance of a Plus, such as Lee tested, would be quite noticeable. I don't quite know how a Plus was hooked up to Ethernet, but the SCSI-based connections reportedly suffer from very poor performance. That could account for the minor speed improvements being touted. Again, Mac II level & up workstations doing work w/ high net traffic will show dramatic improvements switching from LocalTalk to Ethernet. -- Bill Douglass, TCADA "I dreamed I was to take a test, in a Dairy Queen, on another planet." L. Anderson
desnoyer@apple.com (Peter Desnoyers) (10/13/89)
In article <3335@mace.cc.purdue.edu> ajq@mace.cc.purdue.edu (John O'Malley) writes: > MacUser's May 1989 issue has got a great report that compares LocalTalk > vs. EtherTalk, AppleShare vs. TOPS, and also covers performance differences > when using a Mac Plus, SE, II, IIx, SE/30, and Compaq 286. > > It's a great report. Required reading for any Mac network manager, I'd say. Please, this has been discussed before in this newsgroup. The article you mention is flawed in its methods, and reaches unsound conclusions. Peter Desnoyers Apple ATG (408) 974-4469
truesdel@ics.uci.edu (Scott Truesdell) (10/14/89)
xdaa374@ut-emx.UUCP (William T. Douglass) writes: >I don't quite know how a Plus was hooked up to Ethernet, but the SCSI-based >connections reportedly suffer from very poor performance. That could account >for the minor speed improvements being touted. There ya go! >Again, Mac II level & up workstations doing work w/ high net traffic will >show dramatic improvements switching from LocalTalk to Ethernet. Just as I thought, especially the "high net traffic". Even if the hardware is slowing one down, just being able to cram more stuff down the wire is worthwhile in a heavily used network. --scott -- Scott Truesdell