Sorry for delay. See below for my results (WSL/Ubunto): 1000 files: ln -sf * using xargs: clock: 1.58s, cpu: 0.27u, 0.30s * using pair-wise (Perl) clock: 0.10, cpu: 0.04u ,0.09s
10000 files: ln -sf * Using xargs: clock:16.7s, cpu: 3.07u, 3.38s * Using pair-wise (Perl): clock: 0.59, cpu: 0.13u, 0.46s Estimating overhead of using xargs is 10X - about 16 milliseconds per file, vs 1.6 milliseconds vs using single process. The performance difference is 10X, and for many use case, it's far below SLA, which will necessitate a performance driven solution, where I think built-in 'pair-wise' option for ln & mv is much better vs. "do it yourself" ad-hoc script. I also did the same measurement for 'cp' - with a file size of 1K: 1000 files: cp -f * Using xargs: clock: 2.26, cpu: 0.30u, 0.67s * Using pair-wise cp (Perl) clock: 0.15, cpu: 0.05u, 0.10s And repeated for 10k. It actually show that cp (of small file), the pair-wise copy will be more performant vs, xargs cp solution. I assume this has to do with buffering/caching/... For my use case, suppoting just pair-wise on 'ln' and 'mv' will address my problem. Will be happy to see this imprvoemtn coming. Thanks in advance. Yair On Tue, Aug 27, 2024 at 1:01 AM Rob Landley <r...@landley.net> wrote: > On 8/25/24 18:34, Glenn Golden wrote: > > Since you were reporting 2 min, was wondering what your platform is and > > whether there might be something else involved eating the 2 min realtime? > > Slowest exec I've had to deal with on a modern still-active system is mac > homebrew, which was about 1/3 of a second each exec() on fairly powerful > hardware. (It did not LIKE running ELF binaries on a mach-o kernel.) > > Of course "this posix system can't really do shell scripts" means your OS > is > kinda conceptually broken... > > Rob >
