Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!seismo!cmcl2!husc6!panda!genrad!decvax!decwrl!reid
From: reid@decwrl.DEC.COM (Brian Reid)
Newsgroups: comp.sources.d
Subject: Re: "look" utility?
Message-ID: <8509@decwrl.DEC.COM>
Date: Sun, 8-Mar-87 01:36:06 EST
Article-I.D.: decwrl.8509
Posted: Sun Mar  8 01:36:06 1987
Date-Received: Mon, 9-Mar-87 03:00:25 EST
References: <1916@cbdkc1.UUCP> <837@maynard.BSW.COM> <1145@ihlpf.ATT.COM> <846@maynard.BSW.COM> <271@crys.WISC.EDU>
Reply-To: reid@decwrl.UUCP (Brian Reid)
Organization: DEC Western Research
Lines: 83

In article <271@crys.WISC.EDU> solomon@crys.WISC.EDU (Marvin Solomon) writes:
>
>Look is much faster than any of the greps, since ...
>Fgrep is always the slowest ...
>Egrep is always the fastest.

Hi, Marvin. I'm always glad to see actual measured data on the network. But
clearly you don't have the James Woods (ames!jaw) grep. At our site we call
it ngrep. Here are the times for ngrep. Your machine CRYS.WISC.EDU is a 
Vax 780; I have a 785. Therefore I re-ran all of your tests, to avoid having
to convert the numbers between dissimilar machines. Here is my output of your
tests. Note that in every case, ngrep is faster than egrep by a huge amount
in cpu time and by a significant amount in real time.

$ time grep -l a /usr/dict/words
        4.2 real         2.9 user         0.4 sys  
$ time fgrep -l a /usr/dict/words
        0.5 real         0.0 user         0.1 sys  
$ time egrep -l a /usr/dict/words
        0.2 real         0.0 user         0.1 sys  
$ time egrep -l \^a /usr/dict/words
        0.2 real         0.0 user         0.1 sys  
$ time ngrep -l a /usr/dict/words
        0.3 real         0.0 user         0.1 sys  
$ time ngrep -l a /usr/dict/words
        0.2 real         0.0 user         0.1 sys  
$ 
$ 
$ time grep -l jade /usr/dict/words
        3.7 real         2.3 user         0.4 sys  
$ time fgrep -l jade /usr/dict/words
        2.5 real         1.5 user         0.3 sys  
$ time egrep -l jade /usr/dict/words
        1.8 real         0.8 user         0.3 sys  
$ time egrep -l \^jade /usr/dict/words
        2.5 real         0.9 user         0.3 sys  
$ time ngrep -l jade /usr/dict/words
        1.0 real         0.1 user         0.2 sys  
$ time ngrep -l \^jade /usr/dict/words
        2.3 real         0.3 user         0.8 sys  
$ 
$ 
$ time grep -l zygote /usr/dict/words
        3.9 real         2.2 user         0.5 sys  
$ time fgrep -l zygote /usr/dict/words
        4.5 real         3.1 user         0.5 sys  
$ time egrep -l zygote /usr/dict/words
        3.3 real         1.8 user         0.4 sys  
$ time egrep -l \^zygote /usr/dict/words
        3.1 real         1.7 user         0.4 sys  
$ time ngrep -l zygote /usr/dict/words
        1.9 real         0.1 user         0.4 sys  
$ time ngrep -l \^zygote /usr/dict/words
        2.7 real         0.2 user         0.8 sys  
$ 

However, that's just a Vax. If you want to see ngrep really sing, watch while
I run it on a DECWRL's "Titan" machine, which is about 15 times the speed of
a 780:

$ time ngrep -l \^Zurich /usr/dict/words
        3.7 real         0.0 user         1.5 sys

The real time is slower because my Titan has only one disk, and the
disk arms must move back and forth between /usr/dict and /bin/ngrep.
I can reduce the effect of having only one disk arm by using a longer file to
search:

$ ls -l /usr/dict/words
-rw-r--r--  1 root       201314 Nov  5 10:43 /usr/dict/words
$ ls -l /usr/dict/web2
-rw-r--r--  1 reid      2498185 Aug 13  1985 /usr/dict/web2
$ tail -1 /usr/dict/web2
Zyzzogeton
$ time egrep -l Zyzzogeton /usr/dict/web2
/usr/dict/web2
       10.3 real         5.1 user         2.5 sys
$ time ngrep -l Zyzzogeton /usr/dict/web2
/usr/dict/web2
        7.8 real         0.4 user         2.3 sys

Conclusion: be careful with claims about how fast egrep is, since ngrep is
almost always significantly faster.