On Mon, Nov 12, 2007 at 09:40:47 -0600, Brian Moon wrote: > >The script is good, thanks for taking the time. But your results are > >strange (especially the 30% improvement of get method). Here's what I > >get: > > > > $ret = ... void context > > > >trunk 101.255008 102.128995 > > 13.434119 13.074898 > > > >noreply 97.261798 13.326187 > > 12.990013 12.990013 > > > > > >I don't see the overhead of assignment BTW, numbers are roughly the > >same. So my questions are: are you sure patched Perl module was used > >instead of a system-wide one (you probably have to set PERL5LIB)? > >Have you rebuild and restarted the server? > > Actually, I know what it is. I only used the patched client. So, still > the savings is all in the return of the variable. With trunk server and > patched client, I see the same savings as the patched server. > Basically, this all comes down to perl's assignment from a sub routine. > > Try with the patched client and trunk server and see what you get.
When you use patched client and trunk server, you effectively do not wait for the results in the client (even though they are being sent to it). That's why you see the speedup (though the mix would block at some point). No, Perl assignment is not that slow, see numbers above. You see 14->4 saving, and you'll perhaps see a bit more when you use patched server. Everything is correct :). -- Tomash Brechko
