Hardware specs for FM8 server (Mac OS X)

[mhoefer]

Hello

I'm about to upgrade a non profit org from FM Server 5.5 to Server 8. I have 2 boxes I could use as the server (I am not running OS X Server).

Currently it is running on and 800MHz G4 with a simple 2 Disk Raid (Via PCI Card). So that is one choice. The other option is a Dual 500MHz G4. I think I could move the RAID set up to that machine. So I guess the question is would the Dual 500MHz give a signifigant (any?) boost over the single 800MHz.

Unfortunately a server upgrade is not in the cards right now. The org has less then 10 users.

Also have noticed a couple of tips to user 10.3.9 vs 10.4.x is that still current recomendations?

Thanks in advance for any advice.

Mike

[Steven H. Blackwell]

I'd say go ahead with the 800 MHz machine if you don't expect heavy usage. How much RAM do you have installed?

Steven

[mhoefer]

Thanks Steven, it saves quite a bit of work keeping it on that machine.

I think we have 512MB in there now... Seems like a good time to beef that up a bit! I think the max for this model is 1.5GB.

Mike

[Steven H. Blackwell]

Well, I would up the RAM to 1.5GB then. If you're ever going to do serious IWP, that's not enough. But it will work for now.

Be sure to use the latest v-rev of either Server 7 or Server 8, whichever you install. And watch the integrity of the hard drives. At least with the tower model you can swap them out.

Steven

Edited December 25, 2005 by Guest

[xochi]

In FM7/8 server, the server RAM Cache size is crucial. In many cases, it's a bigger deal than either CPU speed OR disk speed.

What is your RAM cache size?

[Steven H. Blackwell]

In FM7/8 server, the server RAM Cache size is crucial. In many cases, it's a bigger deal than either CPU speed OR disk speed

What is the basis of this statement? The hard drive subassembly is the most crucial aspect of a FileMaker Server CPU. You should allocate an amount of RAM to cache sufficient to produce cache hits in the 90 to 95 range during normal usage. As a matter of practice I start at 64 MB and adjust based on the statistics in the SAT Tool.

Refer to the Server Tech Brief on the FMI web site for more information.

Steven

[xochi]

What is the basis of this statement? The hard drive subassembly is the most crucial aspect of a FileMaker Server CPU.

Under FM5/6, since RAM cache was limited, I think that most everyone agreed that in terms of overall performance, the following rankings applied:

#1 Hard Drive Speed

#2 Network Speed / Latency

#3 CPU speed

#4 RAM cache size

Under FM 7/8, I think the relative importance has changed. I don't think a general consensus has appeared. My opinion (for the stuff I'm working on) would be:

#1 RAM Cache Size

#2 Network Speed / Latency

#3 CPU Speed

#4 Hard Drive Speed

Just my opinion, but my own informal test results seem to back it up.

[Steven H. Blackwell]

My own experience with both Server 7 and Server 8, both standard and Advanced, is that the hard drive subassembly remains the core and most critical aspect of FileMaker Server deployment. This is especially the case when considering the new method of backing up and the processing power of newer CPU's. Maybe moreon this later; Iam in transit today.

HTH

Steven

[xochi]

Hi Steven,

I wonder if we are both correct, and our different experiences are due to the nature of the operations we are doing?

One thing that I'm not clear on (because I've not found it documented) is whether the "RAM Cache" is a Read Cache, a Write Cache, or a Write-through-Cache. Mac OS X does a really good job of cacheing disk reads, which means that in some cases (where you are only reading data, not changing it), the size of the filemaker cache doesn't matter that much. The reason presumably is that even though FM hasn't cached the data in RAM, the OS has.

I've also encountered situations where the cache flushing appears to be broken (or poorly documented). I've seen cases where the "flush when idle" setting tries to flush the write-cache during long operations (such as a looping script, or a big replace-field-contents script). In these cases, setting the "flush" time to 15 minutes can significantly improve performance.

FM Server doesn't come with very much in the way of profiling tools. However, I've found that you can learn a lot by simply opening "Activity Monitor" and watching the CPU, Disk, and Network meters, on both the server and the client. If you start a long operation and watch CPU, Disk, and Network, you can usually determine where the bottleneck is with simple logic. For example, if the server's CPU is maxed out (100%), but the Disk and Network are not, the conclusion would be that a faster CPU would help (more than would more/faster RAM or Disk). Similarly, if the CPU is not maxed out, but Disk reads are high, you would conclude that a faster CPU wouldn't help, but more RAM and/or faster Disk might.

Since you seem to feel that FM7/8 still is mainly disk-bound, have you found any differences between FM5.5?