>Now I put my program in a computer equipped with Cyrix MII300, 64MB
>SDRAM, and using Ms WIN95 & WIN98 (ms networks client) to link up other
>computers using 10 baseT.
>
>The Network now have 5 user and the parent table have 6000+ records and
>children table have 12000+.
>
>Now the problem I run the program on the main computer (where the
>program installed), the program function very fast, however when I run
>the program at other computer, the program will work dramatically slow.
>
>Can you suggest any tips to improve the speed of the program??
>
>Any VFP or Ms WIN 95/98 setting will help the speed??
>
>Thank u

You sent me this message via email as well; I'm including the text of the message I sent in reply here, so that others can comment or expand on the reply with their own recommendations. it's far more productive to use messaging here on UT rather than sending requests for assistance via email, unless you're looking for specific actions from a specific individual, like hiring them to take care of the problem for you. Please use UT for further discussions of this problem.

It sounds like one of two things is happening; either the amount of data passed over the network to satisfy queries is very large, so that the network is becoming badly bogged down transferring information, or that the server is getting overburdened servicing local file access requests, resulting in poor performance for other stations trying to access the data.

If most of your application requires non-opmtimizable queries to be processed constantly (lots of reports running, or browses against tables using SET FILTER, or SQL queries where the query can't take advantage of Rushmore optimization, which requires that search characteristics match the structures of the indexes on a table), then you probably have the first situation. The upper limit of the data transmission of Ethernet is in the 300-400KB/sec range; split this among 5 stations, and data is being moved at perhaps 80KB/sec. Compare this to data transfer between the local hard drive and the CPU for local file access, which may happen at 4-5MB/sec, and the performance difference is staggering.

Optimizing your queries to take advantage of Rushmore, using parameterized views rather than filtered sets of tables, and adding some index tags to help query performance all might help.

Another alternative would be to switch to 100BaseT rather than 10BaseT networking. If your network is cabled with CAT5 cabling, replacing your existing network cards and network hub with 100Mbit capable components will give you a roughly 10x boost in performance. The price of 100BaseTX hardware has dropped dramatically in the past year or so; a good quality, low-end 10/100BaseTX , like the NetGear 310TX, runs in the US$35-40 range, and 8 port 100BaseTX hubs like the FE108 are well under US$200.

If you do not have CAT5 cabling, you'd have to recable as well as replace the NICs and hub.

If the workstation performance seems to change depending on how heavily the machine sharing it's drive for use by other stations is being used, you may be looking at service performance issues. There are several directions to go with this type of problem - you may be able to change the foreground and background task allocation of resources to some extent through the Performance tab of Control Panel/System; you could try changing the role of the system to network server, which allocates more time to service background tasks; this will have some impact on local performance, and will devote more processor time and memory to the network tasks.

You might consider adding a machine to use as a server, so that it doesn't have any foreground task responsibilities. You could even do this by adding another system running Win9x, although I'd recommend NT Server. You can see if this will have an impact by not running the application on the current host machine and then seeing how performance is on other stations.

You now are left with issues of coding changes. I haven't seen your application (and I don't have time to evaluate it for free) but if you do lots of queries, use SET FILTER TO a good deal, and perform other operations that require the system spin through all the records one record at a time repeatedly, rewriting portions of the code to take advantage of Rushmore optimization and passing smaller subsets of the data over the wire may help. Evaluating your application is probably a job for a consultant.

Good luck,

Ed Rauh