Re: Perfomance problem backupscript (more details and tests)

"Ruud Baart" <[email protected]> wrote:

> Thanks for the comments on my performance problem. I followed some suggestions and 
> my conclusion is that star is not performing as it one may expect. I hope someone has 
> some suggestions
>
> The reasons:
> 1. As Thomas suggested I timed the performance of find and compared it to star. This 
> shows clearly that star need much more time than find. The results are at the end of this 
> mail. The timing on a real production server is even more dramatic:
>     time find ... directories ... -ctime 1>/dev/null
>     real 7m41.602s, user 0m0.984s, sys 0m28.531s
>
>     time star -c -acl -H=exustar ... directories ..>/dev/null
>     ---->>>>> interrupted, takes too long to wait for <<<<----
>     real 56m52s, user 9m16s, sys 9m50s  
> 2. Joerg Schilling suggested that the debian version was old. So I tested with 
> star debian versions 1.5a57 (sarge distribution) and 1.5a60 (test distribution) and with 
> version 1.5.a65 compiled on the machine. All the same result
>
> 3. I tested with XFS and Ext2 filesystems. Same result.
>
> 4. I tested on different machines (Xeon and AMD). All systems showed more or less the 
> same results.
>
> 5. I tested with kernel 2.6.8-2 (debian kernel) and with 2.6.12-3 (own version). Makes no 
> difference with any of the tested star versions.
>
> 6. I've not tested the tapestreamer(s), but for now I don't see any reason for doing that.
>
> Below the timing results on a testserver.
> ======================================
> These tests were done on: 
> /dev/hde5     ext2    963M  586M  327M  65% /tmp
>
> Kernel: 
> Linux wyatt 2.6.8-2-386 #1 Thu May 19 17:40:50 JST 2005 i686 GNU/Linux
>
> wyatt:/# time find /tmp -ctime 1 >/dev/null
> real    0m1.008s
> user    0m0.012s
> sys     0m0.083s
>
> Star version 1.5a60 (debian testing version)
> ------------------------------------------------------
> wyatt:/# time star -c -acl -H=exustar /tmp>/dev/null
> star: 60272 blocks + 0 bytes (total of 617185280 bytes = 602720.00k).
> real    0m30.459s
> user    0m0.317s
> sys     0m4.836s
>
> time star -c -H=exustar /tmp>/dev/null
> real    0m29.743s
> user    0m0.290s
> sys     0m4.718s
>
> time star -c -no-fifo  /tmp>/dev/null
> real    0m30.325s
> user    0m0.261s
> sys     0m4.396s
>
> gun tar test
> ----------------------------------------------------------
> wyatt:/# time tar -c  /tmp>/dev/null
> real    0m1.054s
> user    0m0.212s
> sys     0m0.110s

If you like to do a _real_ _test_ why do you compare the star time 
against a dummy command?

Why don't you use the right star option (-time) to _easily_ see the
star peformamnce? If you like to use star for more than exremely small
and simple tasks, please start with reading the man page. Star roughly
gives you twice as much as features then GNU tar and you sould know them....

_I_ would not call 20m MB/s (as you would know if you did use -time)
a low speed.


Jörg

-- 
 EMail:[email protected] (home) Jörg Schilling D-13353 Berlin
       [email protected]		(uni)  
       [email protected]	(work) Blog: http://schily.blogspot.com/
 URL:  http://cdrecord.berlios.de/old/private/ ftp://ftp.berlios.de/pub/schily