On Mon, Feb 17, 2014 at 4:25 PM, Phil Smith <philbo...@gmail.com> wrote:
> On Mon, Feb 17, 2014 at 6:16 PM, Charles DeRykus <dery...@gmail.com>wrote: > >> >> On Mon, Feb 17, 2014 at 12:41 PM, Phil Smith <philbo...@gmail.com> wrote: >> >>> I'm currently loading some new servers with CentOS6 on which perl5.10 is >>> the standard version of perl provided. However, I've also loaded perl5.18 >>> and I don't think the version of perl is significant in the results I'm >>> seeing. Basically, I'm seeing perl performance significantly slower on my >>> new systems than on my 6 year old systems. >>> >>> Here's some of the relevant details: >>> >>> + 6 year old server, 32 bit architecture, CentOS5 perl5.8 >>> perl, and in particular regexp operations, perform reasonably fast. >>> >>> + Very new server, 64 bit architecture, CentOS6, perl5.10 (and have >>> tried perl5.18) >>> perl, and in particular regexp operations, perform significantly slower >>> than on the 6 year old server. That struck me as odd right off. I though >>> surely, perl running on a modern high-end cpu is going to beat out my code >>> running on 6 year old hardware. >>> >>> I've compared CPU models at various CPU benchmarking sites and the new >>> CPUs, as you would expect, are ranked significantly higher in performance >>> than the old. >>> >>> I've also installed perl5.8 on the new 64bit servers and the performance >>> is similar to that of perl5.10 and perl5.18 on the same 64bit servers. >>> Given that, I don't think perl version plays a significant factor is the >>> performance diffs. >>> >>> Is it an accepted fact that perl performance takes a hit on 64 bit >>> architecture? >>> >>> I've tried comparing some of the perl -V and Config.pm results looking >>> for significant differences. That output is pretty verbose and the most >>> significant difference is the architecture. >>> >>> I could provide some of my benchmarking code if that would be of help. >>> The differences are significant. The only reason I'm looking at this is >>> because I could see right off that some of my code is taking 30-40% longer >>> to run in the new environment. Once I started putting in some timing >>> with Time::HiRes I could see the delay involved large amounts of regexp >>> processing. >>> >>> Right now, I'm just looking for any opinions on what I'm seeing so that >>> I know the architecture is the significant factor in the performance >>> degradation and then consider any recommendations for improvements. I'm >>> happy to provide further relevant details. >>> >> >> This sounds like it could be something OS-specific and, googling >> "CentOS regex performance" generates hits, eg, >> >> >>> >>> http://pkgs.org/centos-5/puias-computational-x86_64/boost141-regex-1.4.0-2.el5.x86_64.rpm.html >>> >> >> No, I really don't think it is specific to a version of CentOS. I've >> installed various permutations of 32 and 64 bit CentOS 5 and 6. The better >> performance seems to follow the 32 bit architecture rather than a specific >> Perl version or CentOS version. >> > Newer perl regex engines have added Unicode support which can add drag. I'd be surprised though if just the 64-bit architecture itself was totally responsible for major slowdowns. Some of the issues are mentioned here: http://stackoverflow.com/questions/17800112/upgraded-from-perl-5-8-32bit-to-5-16-64bit-regex-performance-hit Per above, some of the items, you'll need to be careful with: were both Perls compiled with the same flags? are both perls threaded perls (disabling threading support makes it faster) how big are your integers? 64 bit or 32 bit? what compiler optimizations were chosen? did your previous Perl have some distribution-specific patches applied? Basically, you have to compare the whole perl -V output -- Charles DeRykus As you can see, you need to be carefully examining the comparison scenarios. -- Charles DeRykus
