There are a couple of issues here...

First the timeouts are configurable, so you can make Web Connection take as long as you need it to. If it takes 8 minutes the Web Server will also time out (by default), so you will need to adjust the IIS request timeout as well.

For Web Connection in COM mode you can also use the SingleMode querystring option to run requests outside of the Web Connection pool. When you use SingleMode (look it up in the docs), Web Connection starts up a new COM server separate of the pool and it has no timeout. It's easy to do, all you do is append the SingleMode flag to the URL on the querystring.

But this is generally a bad practice, because the client has to wait for the response and sitting at a non-interactive screen for minutes will either make the user go away and give up or worse, retry the request.

The right way to do this is via Asynchronous request processing and offloading the query to another machine if the CPU load is heavy for this task. In addition, to freeing the server pool and not tying up IIS threads, you also get a chance to provide feedback as you're doing the search to let your users actually see something happening.

There's a lot more information here which describes this process in great deatail:

http://www.west-wind.com/presentations/wwAsyncWebRequest/wwAsyncWebRequest.htm


+++ Rick ---

>I am facing a situation with the combination of WWC and PHDBase that I am trying to resolve. Before executing a PHDBase search, in here for example when someone does a search, I am applying a SetTimeout() at the Msxml2.ServerXMLHTTP object level, so assuming a search would be too long, the timeout would occur and a message would be returned to the user. So far, so good, when this is rarely happening, it is working fine. The WWC transaction is completed with success.
>
>But, when someone execute a search on all years, I have to run the process into a loop. So, the loop will be executed 8 times. Assuming all results are returned within the timeout, all would be ok at the search level. But, the problem is that because this is into a loop, the total amount of time of each search for those years could exceed the WWC timeout. So, in this case, it generates a WWC failure with the standard message being returned to the user and the WWC VFP server instance is being restarted.
>
>For me, I can't afford to have that to be done. I need to make sure that, if a WWC VFP server instance is started, it would be because I would have decided so.
>
>So, I've been thinking about this PHDBase structure again tonight. Basically, the search is not done on the production table. I wouldn't recommend that to anyone. There are obvious reasons for that. So, I have a robot which is synchronizing additional yearly tables based on that main table which are only used for the search. So, I have a table of messages per year which sums them all.
>
>I thought that by splitting those tables that I would gain on speed result. But, is that really the case? While I am indexing PHDBase on a full copy of the main table, which takes about an hour to do all that because of the million messages, in order for me to compare speed result, I thought about posting a message in here to collect feedback.
>
>PHDBase works with VFP indexes. So, if I have 150,000 messages in average per table for the search compare to having one table with one million messages, if we are using VFP indexes, isn't the speed result be about the same?
>
>Anyone did that comparism before? The only comparism I did so far for that search structure was to use SQL Server. So far, my results gave that SQL Server search engine was as fast as PHDBase. I didn't switch to SQL Server for that search engine because I have something which is working ok for now.
>
>But, I would like to resolve that issue when the search is timing out because of big search process. If I would have everything into one table, which is what I would know in about two hours, if I would hit a big search process, the timeout would occur and the WWC transaction would be returned ok to the user with the message telling that the search criteria could be refined because the word entered is too generic thus too may results. And, again, that would work because I would eliminate the loop process of accumulating the number of seconds for each year causing, in rare cases, a timeout at WWC level.
>
>For those who are wondering why I don't increase the timeout at WWC level, well, I just don't want to. I figure that 50 seconds should be ok to return a transaction. Other than that, we better advise the user about what is going on. Because, when that would be the situation, the user will probably cancel the transaction before it would finish assuming that I would increase the timeout to 120 seconds, for example.