John,
It does sound like a tough one < s >.

First, let me see if I have the most critical problem well understood... you NEVER lose any records, but rather have some records that fail to be indexed until you do a REINDEX. Is that correct?

Assuming so, it does seem odd (though much much better than losing actual records!!!!). Have you noticed if this happens under only certain circumstances, like possibly while or right after someone has printed a report (through the server, I would presume). Let me put it this way... if ONLY the basic data-entry and other non-report functions of the application are used, does the problem still occur?
I think it might also be useful to try to turn OFF write cacheing again, especially on the server. And I have heard of some people backing OFF the XP SP1 'fix' because of problems encountered, so if you have it installed it might be worth considering to remove it.

Now, on to the other problem (which may be the same one)...
I would guess that there is a good chance that somewhere in the application a "dangling object reference" is happening, causing the system to eventually use the SWAP (which will always slow things wayyyy down).
While there are circumstances where these can occur by VFP bugs, I believe the most common cause is programming error. Somewhere on his site ( http://www.geocities.com/df_foxpro ), David Frankenbach has a utility to help you find these, so it might be worth checking that out. It is possible that the grid error on the workstations is related to that too.

good luck


>Hi,
>Thank you for reply.
>
>>>Do you use TRANSACTIONs in this app?
>Ya, we are using TRANSACTION in our app
>
>>>What RAM and what OS is the "server" running?
>DDRRAM 133, APACHER/KINGSTOM, WINXP Pro on server
>
>>>I find it hard to believe that turning OFF disk cache would give you >>errors. You turned it off on all machines?... You turned it off for each >>HD on EACH system?...You turned it off while nothing was running on the PC >>except the OS and its services?
>We ever try to turn off the HD Cache on server and one of the workstation. There is nothing run in workstation as services except Norton antivirus.
>On server side, there are norton antivirus and a service called NETROCKEY (for software access protection) run.
>
>>>Finally, is it possible to try running with only 5 or less workstations >>attached? I believe that 9 may be too close to the limit of peer-to-peer >>capabilities.
>We have a client that using the same application, same hardware and environment with 4-5 workstation at the same time. It doesn't cause any corruption yet, however, we got a complaint about system will start slowing down after used for 7-8 hours even just insert one record to a cursor. Also, the grid on the workstation (POS) may cause error "expression dynamicforecolour/backcolor is invalid" after it run for 7-8 hours.
>
>Any ideas?
>
>Thank you