Network Speed

[Chuck]

We have installed a solution that makes use of some heavy calculations to figure out time intervals. The solution runs on FileMaker Server 5. The network that the files are on is rather large, and they are getting unacceptable speed problems. For instance, filling in a row of fields in a portal is taking on the order of two minutes, whereas filling it in locally (files opened on local hard drive) or on a smaller network, there is no time delay at all. They are on 100BaseT, as are our tests.

One of the things I'm going to try to do is see if I can alter the calculations to work faster. But a clue to the degraded network speed was the IP addresses of the server and the client. The first two octets were the same, but the third was different. Doesn't this indicate that the server and the client are on different networks which are connected via a router, and couldn't this be a possible bottleneck?

The server is a Mac G4. Will alotting more RAM to the FMP Server app help?

Thanks,

Chuck

[BobWeaver]

The number of identical octets in IP addresses doesn't necessarily determine whether there is a router. Old Class B IP addresses would only use the first 2 octets to identify the network and the remaining 2 octets to identify the host (IP terminology for computer). If it is a network with more than 255 hosts, then the first few bits of the 3rd octet are used to extend the host addressing. The number of ones in the subnet mask will tell you the size of network. For example if the subnet mask is 255.255.252.0 (binary 11111111.11111111.11111100.00000000), then the network can handle 1023 hosts, and the actual network address is the first two octets plus the high 6 bits of the 3rd octet. So, it is very possible that the client and server are on the same subnet with no intervening router. Certainly, network architecture and traffic will affect performance. You may have to get the network administrator involved to do some rearranging.

Here is some server performance advice from the Colombre and Price book:

- Run FM server on a dedicated machine with no other applications running (no email, no file serving etc.). Ie., your FileMaker server should not be the catch-all office server computer.

- Never turn on file sharing on the server.

- Select only one networking protocol (preferably TCP/IP) unless absolutely necessary.

- Be sure that the server has a fixed IP address, not a dynamically assigned one.

- Turn off virtual memory

- Don't run screen savers on the server. Just shut off the monitor.

- Set sufficient RAM to produce cache hits of 95 or greater (as shown in the user stats window), under normal operation. FM Server 5 on MacOS can make use of up to 40Meg of RAM; on Windows, 9999Meg of RAM.

[This message has been edited by BobWeaver (edited January 04, 2001).]

[Pengo]

One of the things that I found out while installing a filemaker server was that filemaker server doesn

[LiveOak]

There may be a simpler reason. In head to head tests, some of the complex scripts I have done for a biotech client have about a 100:1 speed difference when running on a 100BasedT network vs. the local machine. A lot of this is the difference between a RAM access at full bus speed (16 bytes wide x 100 MHz = 16MB/s) and 100BasedT speed at approx. 5 MB/s (after overhead).

Two suggestions are to:

1) Rework you time interval calculations to run off-line with the today function update. Try different calculation approaches which may be faster.

2) Use persistent RAM disk program such as RAMBunctious (on the Mac) to run FM server off of a RAM drive (backup frequently & use a UPS!). This gives about a 3:1 speed improvement.

-bd

[This message has been edited by LiveOak (edited January 21, 2001).]

[signal]

We saw a nice jump when we switched to a gigabit server into a gigabit/100base switch. The clients are 100 base but the servers ability to push network data faster (as well as a slightly faster G4) made a significant difference.