Hi folks. I'm thinking of upgrading an FMS server, a physical machine that I own and administer, so I wanted some input on whether FM Server is going to benefit from more cores, as for example found in AMD's Ryzen CPUs or whether I should instead opt for less cores but with more performance per core (Intel). Does it utilize threads extensively when many clients are issuing requests, and if yes, has it been like that since, say, version 12 or is it just a recent development?

And also, is this behavior similar in both macOS and Windows or are they different?

Edited July 2, 20196 yr by Buckie

The behavior is similar on macOS and Windows.  Getting more cores is typically cheaper on the Windows side of things, high-core machines on the mac side typically have a lot of extra hardware (GPU,...) that FMS does not need and that drive the overall price up.

As to whether you will benefit from more cores: the only answer can come from your current performance baseline.  Mainly your stats.log and to some extent the TopCallStats.log.

Any server has 4 potential bottlenecks: processing power, disk i/o, memory and network throughput.  In the FM world it is typically processing power and disk i/o, depending on the total user load, the nature of the solution and the design/architecture of the solution.

If processing power is not your bottleneck then you won't get benefit from adding more.

FMS can take advantage of more cores depending on the type of activity.  If you use PSoS a lot, or server-side schedules, or WebDirect or any other kind of server-side activity, then absolutely.

Here's a good recent discussion about processing power for FMS18:

https://community.filemaker.com/en/s/question/0D50H00006tgeVdSAI/fms18-any-definitive-documentationevidence-of-multicore-use

The shot version: figure out from your stats.log where your bottlenecks are, add more resources to the bottleneck area.

Thank you for posting that link @Wim Decorte. Some how I had missed that discussion. Excellent info.

Hi Wim.

At the moment I'm bottlenecked by pretty much everything as it's a very old machine. By upgrading I meant completely replacing the hardware.

So I'm looking to futureproof it a bit, all other things (disks, RAM) are pretty straightforward and can be taken care of later, but you cannot swap a CPU easily. That's why I wondered if it made sense to go the cheaper route with AMD which probably only makes sense if the software can actually efficiently parallelize the tasks.

FMS has been multi-threaded for ages.  FMS18 introduces a boost in better parallel processing provided that the right circumstances are there, see that link I posted for an overview of what is involved.

Either way you need to pick the number of cores and speed of the cores based on what your solution needs and its projected growth.  It's virtually impossible to give you any kind of recommendation in the abstract because there are so many factors at play to pick the right configuration.

Thanks for the advice! I have a better understanding of where to go forward now.

6 hours ago, Buckie said:

So I'm looking to futureproof it a bit, all other things (disks, RAM) are pretty straightforward and can be taken care of later, but you cannot swap a CPU easily. That's why I wondered if it made sense to go the cheaper route with AMD which probably only makes sense if the software can actually efficiently parallelize the tasks.

CPU's can be swapped out easily too. Create a virtualized environment. Depending on your requirements, you then also have the option of creating a cluster of virtual hosts and load balance servers across the cluster. If one dies, the performance might drop, but you're still up and running. And adding CPU (or any other) resources to VMs is trivial. Plus, you can create backups of your entire VM and load it elsewhere quite easily. Upgrading your FMS then also becomes much less of a problem, ie I've got a Windows 2k8R2 server running FMS, which needs to be upgraded to Win2016 for me to be able to run v18. Using a VM I can simply install a new server within my environment, build it when its convenient, transfer DB's for testing and when I'm happy I can simply replace the old server with the new on a cold winter night.

If I had a physical FMS server, I would need to purchase new hardware or work very hard over a weekend to ensure everything was back up and running come Monday. I could do that because the business I work is closed over the weekend, but that's not true for a lot of businesses. 

In a clustered virtual environment, you could add a new host with heaps of resources, migrate all your VM's across and no one would be the wiser. Then you decommission your old host(s).