 | |
#1
 
| |||
| |||
|
Performance Questions -
07-31-2003
, 02:20 PM
#2
| |||
| |||
|
|
I am designing a retrofit for an existing FileMaker installation. Today the company is using FMPro 5 (not server), and accessing the files over a 100base-T file server via a windows 2000 machine, which then hosts it for other clients. Performance is not stellar, and they have a non-trivial number of crashes and corruptions. They have a reasonable budget, but price is an issue. I'm a Mac guy, and they are not platform snobs, so I'm suggesting the following setup: Power Mac G4 (1ghz or so) running OSX 10.2.6 and FM Pro Server 6 Two IDE drives formatted (using Disk Utilities) as RAID 1 Client machines (a mix of Mac & Windows) running FM Pro 6. FMServer will be set to auto-backup a rotating set of daily backups. We will handle offsite backups by some other method. Questions: (1) We are also building a CDML-based web system for internet access. Everything I've read says "Don't run FM6 Unlimited on the same machine as the FM Server" but my experience has been that this works just fine. In fact, from a networking point of view, I would predict that having FMU serving CDML from the same machine as FMServer would INCREASE performance, beacuse it's always going to be faster to read data from localhost than over an ethernet network, no matter how fast. (2) I'm debating using Apache and the web connector as well. Some info says it only runs under Mac OS X SERVER not regular consumer Mac OS X. What's the truth? (3) If I do choose to run everything from one machine (FM Server and FM Unlimited) would having an older 2-cpu machine perhaps beat a newer single cpu machine? Although FMServer isn't threaded for multi-cpus (?) I'm assuming that Mac OS X would properly allocate the CPUs to both FMServer and FMUnlinmited as needed. Tips / experiences appreciated. -- To send email, remove the invalid and nospams. |
#3
| |||
| |||
|
|
FMS 5.5 (the latest version-- there is no FMS 6) currently contains the ability to corrupt files in certain conditions-- apparently some manual shutdowns. This issue has been around now for a year, but remains unrepaired by FMI. For that reason, OS X is not currently the best choice for FMS. FMI does not recommend Mac OS 9, and that leaves Windows 2K Server or 2003. Having said that, I have not yet experienced that file corruption on two systems running FMS 5.5v2 on OS X, but your mileage may suck. There have been some reports of successful Terminal workarounds. |
happen even when the Server pauses the files for backups?) It's a good
point though, as high reliability is a key feature.
Quote:
|
There are a lot of reasons not to run FMU on the same box as FMS. The easiest is that you will hurt the performance of FMS, which needs all the CPU and, mostly, drive/controller attention it can get. |
FM6U is using a database hosted over the network interface, why would it
ever be hitting the drive? (Unless FM6U caches data locally on the
drive, which would be poor design...but of course that's never stopped
Claris in the past :-)
Quote:
|
In addition, if you do not place FMU on a separate drive from FMS, you will open up FMS to file sharing; even if you do, you increase that risk. |
FMS and FMU would try to make the files openable over the network? Or
are you talking about FMU accidentally opening the file directly from
the hard disk (instead of through FMU)?
thanks again
--
To send email, remove the invalid and nospams.
#4
| |||
| |||
|
|
John Weinshel <john (AT) datagrace (DOT) biz> wrote: Both FMS and FMP (FMU is basically FMP) are drive intensive and drive bound. As mods are made to the hosted (on the web) files, those mods and additions need to be written to disk. That activity is taken away from the FMS disk activity. Not sure I agree here. Here's my reasoning: Let's say a user is editing a file that is hosted via FM Server. FMU tells FMS to make the change. FMU doesn't do the disk writing, FMS does. In which case it should be faster for the message "update these records with these values" to transfer on a local machine (localhost) than over the actual network, even if it is 1000base-t. In other words, I'm claiming that there is absolutely no need for FMU to ever touch the disk (on the machine that is running on) if it's using a file hosted over the network. And if FMS is doing all the disk access, it shouldn't be affected by having FMU running on the same machine (not counting of course, CPU resources). FMU must be set to multi-user and its volume open to sharing, in order for it to receive the files hosted by FMS. FMS must absolutely live on an unshared volume. Therein lies a potential conflict. I don't agree, but perhaps my setup isn't clear. 1. FMS is running on one machine (that has file sharing turned off) hosting my 5 database files on my local network. 2. FMU is running on the same machine, and it has those files opened VIA THE NETWORK (i.e. it is being served these files from FMS). The files are opened in "Single User" mode. Neither FMS nor FMU seem to notice or care that they are both on the same machine. To be clear: I did NOT open the files in FMU directly from the disk, I used the HOSTS button and typed the ip address of the same machine. 3. Web companion is turned ON for FMU. I've got this setup working right now, and I can't see any problem with it. I've just tested from another machine on the network, I can access my database via CDML by pointing my web browser to it, or I can open it from within FileMaker Pro by using the Hosts button... As far as I can tell, FileMaker Server is doing the database hosting, and FMU is doing the web hosting, and it all works! I neglected to mention a third conflict-- the network hardware and software. FMS is very chatty, and cannot bind to more than one NIC. Now you want to share that computer's network resources with a client, FMU. More conflict. But aren't all the ports being bound different ones? In any case, it is working fine for me. I suppose there may be some disaster lurking but I don't understand what it might be... -- To send email, remove the invalid and nospams. |
#5
| ||||||||||
| ||||||||||
|
|
Mike D wrote: "In any case, it is working fine for me. I suppose there may be some disaster lurking but I don't understand what it might be..." It isn't that there will necessarily be some disaster; the fact that 'it all works!' doesn't mean that it is a good way to deploy these applications. |
good system. Jumping through uneccesary hoopes because some "wise old"
developers think it's "the right thing to do" is fooey. This is
software, not black magic.
Quote:
|
When we start out as FMP developers, we are at first excited when some seemingly complex operation actually works. But that beginner developer is not worth much; a good developer makes stuff work as quickly as the software and budget allow, |
and researches alternatives. It's amazing how much bad advice there is
out there!
Quote:
|
which means they have to understand the way the software works. |
Quote:
|
What you are proposing might work, |
Quote:
|
but it does not reflect the way the software is designed, |
that they have the principles of isolation and interoperability. In
other words, if FMU can't tell that it's running physically on the same
machine as FMS, then there is no problem.
To illustrate: Turn on personal web sharing (Mac OS X / Sharing panel)
and then enter http://127.0.0.1/ in a web browser. What's this? Cilent
and Server on same machine. Magic? No...the reason this works is that
the TCP/IP stack was designed well.
Quote:
|
and it will not work as well as it should. |
specific?
Quote:
|
Your original question seemed to be 'what is a good way to set up this system', and I have told you. |
Quote:
|
You seem to have your own preconceptions about these products, |
solutions for many years.
Quote:
|
and they are not all correct. |
Quote:
|
If you are unwilling to accept seasoned answers to your questions, you will no doubt learn in your own way. |
dollars and use less electricity!
--
To send email, remove the invalid and nospams.
#6
| |||
| |||
|
|
and it will not work as well as it should. You keep saying this, but I'm finding it to be false. Can you be more specific? |
http://www.filemaker.com/support/tec...hnicalarticles
which gives 8 alternate scenarios for hosting files in various
configurations, some with RAICs, some without, some with FM Server, some
without. NOT ONE recommends the deployment of FMS and FMU on a single
machine. (the article is still current for FM 6 Unlimited with FM Server
5.5, since that part of the product has not changed significantly in the
rev)
In many FM technical articles, white papers and KnowledgeBase articles
(available for a little clicking on the FM site) dealing with the
deployment of FMServer, it is explicitly stated that the recommended
deployment is on a dedicated machine which runs a minimum of other
services, meaning only those services necessary for the system. No web
servers, no mail servers, no file servers, no nothing. Read the FM white
paper on deployment of FM Server on the same page.
Experience deploying a lot of FM solutions tells us that this advice is
good. One of the primary causes of poor performance and poor stability
of FM solutions is improper deployment on servers and networks. If you
want to ensure that your solution is maximally robust and that you've
provided your clients with the best possible advice, follow the
recommendations. You're right, there's a lot of bad advice out there.
But the recommendations that come straight from FMI aren't bad.
BTW, I feel that the FMI documentation is nowhere near explicit enough
in warning about various poor deployment scenarios. They kind of gloss
over what happens when the recommendations aren't followed, seeming to
feel if they just don't mention it, nobody will think of doing it that
(wrong) way. Ostriches, IMO.
"It works in FM, so why shouldn't I do it?" is not sufficient reason for
us to go against FMI recommendations. For example, although it is
perfectly POSSIBLE to access FM files using the OS filesharing instead
of Open/Hosts, we know that this access method leads to instability,
crappy performance, the frequent introduction of file corruption, and
the appearance that the FM software itself is a piece of shit. But
there's no really explicit warning of the dire consequences (which many
of us here have been asked to clean up after, many times) if the
UN-recommended but physically possible methodology is followed.
I could give you other examples of foolish things FM allows, but which
aren't optimal for excellent functioning, but I think you get the idea.
In some OS's, particularly on the Mac, one cannot run both FM Server and
FM Pro or Unlimited on one machine. The Server which is started first
will bind to the NIC, and the guest app, whether Pro or Unlimited,
cannot also bind to the NIC and run guest/host traffic through the same
NIC. In some Windows/NT/XP systems, it is apparently POSSIBLE, even
without dual NIC cards. Even if two NIC cards are present though and you
somehow get each service/app to bind to a different card, you're running
all the activity, netword traffic and read/write actions on one disk. FM
Server wants those resources. Anything which sucks up those resources
will interfere with performance.
Really, in the case of mission critical data, in this case "saving some
money" by trying to run both apps on the same machine is penny wise and
pound foolish. FMI has no self-serving interest in telling you to run
them separately, since you've already purchased both products from them,
and they don't sell hardware.
Hope this clarifies a bit. In FM, as in many areas of life, just because
you CAN do it, doesn't mean you SHOULD.
--
Lynn Allen Allen & Allen Semiotics
FSA Associate Filemaker Consulting & Training
lynn (AT) semiotics (DOT) com http://www.semiotics.com
#7
| |||
| |||
|
|
Lynn allen <lynn (AT) NOT-semiotics (DOT) com> wrote: Mike D <mdNOSPAM (AT) xochiNOSPAM (DOT) com.invalid> wrote: and it will not work as well as it should. You keep saying this, but I'm finding it to be false. Can you be more specific? How about the article on Configuring Filemaker 5 Unlimited at http://www.filemaker.com/support/tec...hnicalarticles To be fair, Yes, this paper does not describe my setup. However, to be more fair, one must also point out that this paper describes "RAIC" as one of the "best practices" to make up for FMU's abhorrent lack of threading (this is 2003, no?). Best practices from a company that designs their software with a 40MB maximum RAM cache and no threading? I suppose they made lemons out of lemonade but no reputable databse/software engineer will say RAIC is the way to solve the threading issue... In some OS's, particularly on the Mac, one cannot run both FM Server and FM Pro or Unlimited on one machine. The Server which is started first will bind to the NIC, and the guest app, whether Pro or Unlimited, cannot also bind to the NIC and run guest/host traffic through the same NIC. Aha! Now we are getting somewhere. This is the kind of technical detail I was looking for. Some meat I can sink my teeth into. Unfortunately, it is false. To rehash my experience: I am successfully running FMServer and FMU on the same machine! FMServer is serving database files on our local intranet, while FMU is serving CDML/HTML pages to the internet. It works, and so far, performance is just fine. On the off chance that the network packets were making a round trip to the router and back (which would negate my claims of better performance) I actually did some sniffing. 'arp -a' confirmed there was no route to self (other than localhost). To be absolutely sure, I unplugged the ethernet cord, fired up Safari, and browsed the CDML pages locally. Worked fine. Therefore, the FMS <-> FMU packets are not actually going over the net. Really, in the case of mission critical data, in this case "saving some money" by trying to run both apps on the same machine is penny wise and pound foolish. Again, you are dodging the issue. You claim it "won't work" but I have a system that is working. The beauty of falsification is it only takes one case to prove a theorem wrong. Q.E.D. FMI has no self-serving interest in telling you to run them separately, since you've already purchased both products from them, and they don't sell hardware. True. My guess is that there are probably some people who don't understand TCP/IP (perhaps including FMI itself based on some of the stuff I've read from them) who could not get it to work. FMI have a vested interests in avoiding tech support calls, etc.. It is not suprising they would issue such pablum given their market position. Hope this clarifies a bit. In FM, as in many areas of life, just because you CAN do it, doesn't mean you SHOULD. A philosophy fitting for our Orwellian times, I'm afraid. (Aluminium tubes? WMD? What Deficit?) Watch Fox News and be happy! The truth is, that I've found at least one exception to their "rule". it is an exception that will save my client $1000s of dollars over the long term (electricity is quite expensive where we are located). Bottom line: One can run FMServer and FMUnlimited on the same machine under certain real-world circumstances. If you want advice on how to do this I'd be glad to help (but beware! it is apparently forbidden knowledge...the high preists of FM Lore may smite you!) :-) -- To send email, remove the invalid and nospams. |
#8
| ||||
| ||||
|
|
Lynn and I have offered you good, time-tested advice. |
false. Do you dispute this?
Quote:
|
If you follow it, your project will perform better, which, generally, means faster |
the gain is marginal...see below).
Quote:
|
and with fewer opportunities for problems than it will if you persist to deploy in the way you describe as 'working'. |
continue to imply that it doesn't work, in the face of evidence to the
contrary, just makes no sense, and suggests you may not understand FM
networking, or TCP/IP networking in general as deeply as you think.
Quote:
|
To equate that advice, supported by our peers' experience, with some sort of political ('Aluminum Tubes', 'Orwellian') or personal ('high priests of FM Lore') bone you appear to want to pick, does your clients a needless disservice. |
personally, but you did start the personal attacks first by calling me a
"beginner"...
Bottom line is, my setup works. Now, as to the whether it will have
reduced performance, this is a question open to empirical testing:
I've done some brief testing under both scenarios (FMU on same machine
as FMS, and FMU on separate machine). The results suggest a slight (10%
or so) performance benefit of the two-machine setup (although take that
with a grain of salt because I didn't have two identical machines to
test on, so I'm extrapolating based on relative CPU speeds of the two
machines).
Again, I thank both of you for the advice, but I hope that you've
learned something here. Next time you are asked whether it's possible
to run FMU and FMS on a single machine, I hope your answer will be
slightly different: "Although FMI doesn't recommend it, some users
have tested it and found it to work, albeit with a possible performance
penalty depending on whether the processes are CPU or Network bound.
You should weigh the performance issue against the cost savings in not
needing a second dedicated computer."
--
To send email, remove the invalid and nospams.
#9
| |||
| |||
|
|
Again, you are dodging the issue. You claim it "won't work" but I have a system that is working. The beauty of falsification is it only takes one case to prove a theorem wrong. Q.E.D. |
recommended deployment for optimum performance and stability."
Don't snip my actual words and then re-interpret them to make yourself
feel better. Many things which are not recommended, and which have been
shown to my experience and that of many other developers I respect to
damage performance and stability, will in fact work in Filemaker. I gave
an example, which you also snipped.
For my clients, I put them into the recommended configuration, and no
other. If you feel comfortable going against FMI recommendations and
those of other experienced developers, so be it. Hang your ass in the
wind as you please. I do know of other developers who run both FMS and
FMU on the same machine. Annecdotally it works. Off & on. Then they
write in to the FSA list saying "why am I having this problem?"
But don't tell your clients you're following recommended practices,
because you aren't. Some clients would rather save a bit of money and
take their chances. Ok. As long as they're making an informed decision.
--
Lynn Allen Allen & Allen Semiotics
FSA Associate Filemaker Consulting & Training
lynn (AT) semiotics (DOT) com http://www.semiotics.com
#10
| |||
| |||
|
|
Lynn allen <lynn (AT) NOT-semiotics (DOT) com> wrote: Mike D <mdNOSPAM (AT) xochiNOSPAM (DOT) com.invalid> wrote: Again, you are dodging the issue. You claim it "won't work" but I have a system that is working. The beauty of falsification is it only takes one case to prove a theorem wrong. Q.E.D. Actually, I said "it will work under some circumstances, but is not the recommended deployment for optimum performance and stability." |
Quote:
|
For my clients, I put them into the recommended configuration, and no other. If you feel comfortable going against FMI recommendations and those of other experienced developers, so be it. Hang your ass in the wind as you please. I do know of other developers who run both FMS and FMU on the same machine. Annecdotally it works. Off & on. Then they write in to the FSA list saying "why am I having this problem?" I'll take your word for it, but I still can't think of any reason why FMU (once set up and workign) would be less stable running on the same machine as FMS than on a second machine. |
have a point. However, 6 motnhs of real world experience saw a
machine not perform as well as the paired machines running the same
software. FMU 6 seems to have a couple of stability problems at
the best of times - running FMS and FMU on the same machine
compunded them. It was only a test machine (my desktop one actually)
but under v light loads, crashed at least once a week. I cannot
recommend using the setup in a production environment.
Quote:
|
Sounds like you have more trust in FMI than I do? I've been continually amazed at FMIs ability to make what I see as such simple mistakes (both in the function of their software, and in their recommendations for its use). |
no
machine should be setup that way (databases stored in the web
directory).
However, and I really do speak from experience, you *will* get weird
crashes if you insist on running FMU and FMS off the same machine.
With the possible resultant file corruption. And my data is waaaay
more valuable tha nthe particular machine it runs on. Thus, *I* will
never run in that configuration.
YMMV
Webko
btw, nice spam block. Some one who does it right. So for once, I will
only post, rather than post and mail :-)
 |
« Previous Thread
|
Next Thread »
| Thread Tools | |
| Display Modes | |
Linear Mode
| |
Powered by vBulletin Version 3.5.3
Copyright ©2000 - 2012, Jelsoft Enterprises Ltd.
Copyright ©2000 - 2012, Jelsoft Enterprises Ltd.