Hi,
Given that filmmaker is unbelievably chatty with the server, did someone tried low latency switches in a LAN with Filemaker, and so some improvements in performance ?
If so, what what the switch model.
I have a colleague who has tested in the opposite direction:
Periodically he does testing with network utility that allows him to simulate high latency conditions and/or low bandwidth conditions. That testing has been very useful to confirm the very strong correlation between latency and FMP client performance.
Which, though it does not answer your precise question, it does suggest to me that you are thinking in a good direction.
2 Likes
Yes, latency is so important for good FMPro performance.
Other ideas for testing:
- benchmark script performance using PSoS
- test using FileMaker Pro installed on the same machine (e.g. localhost)
- test LAN on ethernet (gigabit)
- over the WAN
Generally, we find performance degrades from 1, 2, 3, and 4, with 1 and 2 generally being very similar, 3 is a little worse, and 4 is much much worse.
1 Like
I'm sure latency even on LAN is very important, but was wondering if using better switches vs standard ones provide any speed benefits
Good question.
In theory: I suspect it doesn't matter much as long as your switches aren't broken in some way. As an example, I just tried PING on localhost vs. my desktop (via gigabit ethernet and two switches away):
Localhost: 0.138 msec
LAN: 0.741 msec
so localhost has about 1/5th the latency (relatively).
However, the absolute difference is pretty small - about a half a millisecond. This means in order to add one second of delay, FileMaker would have to send/receive about 2000 messages.
In Practice
I did the same operation (looping through 1000 records and updating them) on both machines:
Localhost: 35 seconds.
LAN: 38 seconds.
Unfortunately this test is not exactly like-for-like:
Localhost: FMP 20.3.1 on a M1 mac mini
LAN: FMP 20.3.2 on an Intel Core i9
My conclusion: once you are on LAN, the latency of your switches is probably much less important than other factors such as CPU speed, software version, etc.
1 Like
in a 'normal' usage, the LAN (speed) is not a problem - compared to the situation where the server is hosted somewhere and You are connecting via celluar services.
If You got 0.2 or 0.9 milliseconds, thats not a problem - until there are hiccups and suddenly the speed exceeds 50 or 100 ms. Even 50ms might be not so problematic..
But if somebody prioritizes some other net-services like MS-Teams, it can/will be problematic. We got customers where the IT prioritizes the 'phone-services (teams..) to avoid hiccups in conversations - and the performance with FMS dropped dramatically...
This is very good example for the reasons Claris/FMI have been saying for years that FileMaker Server should be alone on it's server. I understand a Windows Server is not cheap, some customers do want to control servers' cost.
1 Like
More testing. I repeated my prior test on the LAN but this time using Network Link Conditioner to add artificial 1 through 10 msec delays (on each direction, so 2 through 20msec total ping).
I also did the same test run using PSoS (Perform Script on Server) which presumably has close to zero latency.
Results:
PSoS: 10 seconds
Localhost (0.1 msec ping): 35 seconds
LAN (0.7 msec ping): 38 seconds
LAN (2 msec ping): 82 seconds
LAN (5 msec ping): 161 seconds
LAN (10 msec ping): 218 seconds
LAN (20 msec ping): 557 seconds
Conclusion: latency in FileMaker is an absolute killer.
Here's a graph of the results:
This graph suggests the raw processing takes about 15 seconds, and the rest is network latency, with each millisecond of latency adding about 25 seconds to the total time.
4 Likes
Thanks Xochi,
My LAN is at 0.4 ms
but I was wondering if some high end switches could do better
I've been "blessed" with at least three situations where the LAN is so inconsistent that just cabling, including in wall/ceiling/floor is the problem (all three that come to mind are relatively new construction).
My most recent taste of this was at a client's last Thursday. Staff asked me to run a task for them that I often do as we're still observing the features/performance of a certain complex script. The server is on the 7th floor, I typically run the script from my office (outside consultant, but they provide me an office) on the 6th floor. The script takes about five (5) seconds to run, perhaps a bit more if the set is larger. If I run it on the server itself, using FMP client, it takes about 3 seconds. Last Thursday I showed the staff who asked for this help, the button that calls the script and asked her to do it (which she'd prefer anyway). Her desk ON THE 7th FLOOR is about 2.5 meters from the server booth. The script took at least 30 seconds to run.
Other workstations have a similar effect on simple tasks. In Bahrain we had a similar problem with one area where one LAN port would typically timeout on complex tasks, so we tried the one next to it. It was better but still so-so (slow), and then we cleaned it with alcohol and it was smoking fast. First port was not salvageable. Brand new construction under MY supervision, but not watching every minute of install and cabling conditions around sharp corners and such. I had the contractor replace several such in-wall/floor cables of the two-floor, 1600 sq meter facility.
Not scientific performance tests, but real world issues. We can write snappy code, use fast hardware, etc. but it still boils down to factors often beyond our control.
Sorry for the somewhat off-topic ramble.
2 Likes
And with small clients, things are even harder to control. Only a few months ago I was on-site at a client's office and discovered a stretch of CAT 3 cabling on the LAN.
2 Likes
In Bahrain we installed Cat 8 cable, in Bangkok it’s Cat 6 (if lucky). There’s still some Cat 5/5e around; plus whatever crappy cables they may have at workstations. But in Bangkok the server rooms/booths are typically a horrible mess of spaghetti, dangling switches, insufficient cooling, no dedicated fire detection and suppression, etc. Fun times.
People ask me why I design so much redundancy in backups… I just point to the actual server room/booth (or a picture of them). They nod knowingly. “TIT” This Is Thailand. 
1 Like