garyb@gaboon.UUCP (Gary Blumenstein) (04/23/91)
We are faced with the decision on whether to buy a Sun 4/490 server or get two or three Sparcstation 2's with multiple 1GB SCSI disks. Right now we use three Sparc 1+'s, one Sparc 2, and a 3/280 as general purpose servers on our network. It is not our intention for the 4/490 to *completely* replace our other servers, rather we are looking to add a platform from which we can provide additional disk storage, memory, and cpu speed to our network. In short, we are looking for a platform which will allow us to expand into the future and as our needs dictate. Basically, I was hoping to get some feedback on which configuration would make the most sense to you and why. Since there are so many issues that can influence this decision, I'll start by including a list of pros and cons that I've come up with. Your responses are welcome even if you don't bother to read any further however, the list will give you a better idea of what we are trying to accomplish. If there's enough interest, and the topic hasn't already been beaten to death, I'll summarize for the net or be glad to email the responses to you. If you ask me to, I'll keep your correspondence strictly confidential. One note: We cannot consider another vendor such as Solbourne or Auspex at this time. Sorry guys :-( Our Needs: As previously mentioned, a multi-purpose server platform which provides: NFS services, NIS, mail, tape, and print hosts, and a sometimes compute platform for people with PC's and dumb terminals. Pretty much standard fare distributed Unix shop stuff! Here are some reasons in favor of using Sparc 2's: o It seems to me that multiple Sparc 2's would buy us more aggregate compute power than a single 4/490. While the 4/490 provides greater i/o capacity, by carefully distributing applications across two or more smaller servers, the system could be load balanced to keep i/o activity within acceptable performance bounds. o For us, complexity of administration is not an issue. We are making effective use of NIS and the Automounter to distribute resources across the network. o Most of our network bandwidth is consumed by NFS traffic (largely estimated but some benchmarks have been taken with nfswatch and nfsstat). Currently, our NFS traffic is not very bad at all (thanks to the Automounter) IOPS on our busiest server is in the neighborhood of 20-30 max. o One of the big plusses for the 4/490 is the optional Prestoserve board, but I understand Legato now makes them for the SBus. If NFS traffic increases measurably, we could add Prestoserve's to the Sparc 2's. o Sparc 2's are cheaper and can always be relegated to a user's desk when they become obsolete as servers. o Cost of SCSI -vs- IPI. Since we'd be paying about 3 times more for the same number of megabytes, it seems more cost-effective to stick with relatively cheap SCSI disks. (It seems we're *always* looking for more space!) o We do not yet have any *real* performance monitoring statistics to warrant the absolute need for a high-end compute server. o I am afraid the 4/490 will turn into an "iron alligator"; eating up corporate funds to keep it happy. Most peripherals seem [dis]proportionately more expensive when going to a centralized server. o Our management expects to replace one of our Sparc 1+ servers after getting the 4/490. Even though they agree that our goal is not to totally centralize our operations, I still feel that we will loose some degree of flexibility by reducing our capacity to distribute computer resources, and reduce our ability to "cannablize" equipment from other Sparcs. Here are some points that favor the 4/490. o Sun sales is offering us a _substantial_ discount for trading in our old 3/280 server. This lowers the base price of the 490 to the point where it almost seems like the logical thing to do, if not for any other reason than to stay current with newer technology. Also, we are afraid that we will become stuck with the 3/280 when Sun no longer supports it. o Maintenance on the 4/490 is comparable to the maintenance we pay for our 3/280. o There is some chance that we will need an on-line database sometime in the future (When? Nobody here knows!). It would not be for transaction processing, but rather the need to access large, finite data sets for engineering oriented applications. The high (6Mb/sec) transfer rate of the IPI disks would seem better suited for this application than SCSI. o We have heard rumors that Sun will be using the 4/490 as it's sole platform for newer and more advanced computing ie: multiprocessing. If this means that the 4/490 will be upgradable, it would make good strategic sense for us to purchase one. Some additional questions: Since we are concerned about gaining the most utility from the 4/490, what would happen if for instance, we had set up the 4/490 as an NFS server and later on, our users wanted to run simulations on it? Would this platform provide enough capacity to play multiple roles? Roles that might otherwise dictate segregating in order to provide acceptable performance? Summary: Even though the article may seem biased, I am not terribly concerned which way we go. For us it seems like a win-win situation. I realize there will be some trade-offs in either case. What I'm looking for is your opinions, to see if you agree or disagree with this analysis, or come up with other questions (or answers) that I failed to consider. Thank you for your time and attention. -- Gary Blumenstein United Parcel Service, Research & Development garyb@gallium.UUCP ...uunet!philabs!crpmks!gallium!garyb