Re: [Gnu-arch-users] Re: tla library-add - high cpu

[Tom Lord]

    > From: John Goerzen <address@hidden>

    > > Rob is running on a 64-bit machine.  Patching is a fairly
    > > regexp-intensive operation (due to computing inventories -- those
    > > =tagging-method regexps).  The default DFA and NFA cache sizes of Rx
    > > is measured in bytes -- but allocated in multiples of the machine's

    > I'm working with some large trees (more than 10,000 files) on x86.
    > Would increasing the cache size help me here?

It depends more on the complexity of your =tagging-method regexps than
on the number of files, although the costs of a too small regexp cache
do get multiplied (roughly) by the number of files.

The easiest way to find out is to try it and see.   

Given the ability of users to add to the regexps, perhaps I should
should either just raise the default in tla, or add a .arch-param to
set it, or add an =tagging-method directive to set it, or come up with
some definition of when the cache-miss rate is bad enough to warrent
raising the cache size automatically, or add a heuristic that sets it
based on the lengths of the various regexps.  Hmm.   

Well, let us know what you find.

(And yes, I know I sound like broken record saying this but the
inode-signature foo, which will probably make a much larger
difference, is nearing the front of my queue.)

-t