Subj : multithread question
To   : netscape.public.mozilla.jseng
From : Kumar Iyengar
Date : Tue May 11 2004 05:25 pm

I am in the process of evaluating Spidermonkey and have written a prototype
that embeds Spidermonkey in our application. I have followed the rules(as
far as i can tell) of multithreadedness as Spidermonkey recommends:
*) Compile the library with JS_THREADSAFE,
*) link with the libnspr4.lib file
*) In code call BeginRequest and EndRequest in all places where there might
be a potential block.

Also, i have gone with the model currently where i have a Runtime for each
thread. I know i can go with one Runtime but i have chosen this just for
now.

I have a script that adds 2 numbers. To get the numbers it makes a native
call and another native call to set the result of the addition. In both the
native calls, the work is done between the BeginRequest and EndRequest.

When i run the test with one thread the numbers look good. However when i
run the test with 5 users( i.e. 5 threads) the numbers come out 20 times
slower!. With 5 threads I would have expected a worst case of max 5 times
slower.

Can anyone explain if there are other things to be aware of. Is there a lot
of mutexing in either the js32.dll or libnspr4.lib that causes this?

--kumar iyengar

.