[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: Review Rules and staticly linked packages agains dietlibc

j w r degoede hhs nl (Hans de Goede) writes:

>> user-time:
>>   res.ds.*      res.diet.*      res.glibc.*     ...
>>   0m0.195s      0m0.305s        0m0.340s        ...
>>   100%          156%            175%            ...
>> sys-time
>>   res.ds.*      res.diet.*      res.glibc.*     ...
>>   0m1.406s      0m1.686s        0m2.090s        ...
>>   100%          119%            149%            ...
> Yes, all very interesting but you're using a static inetd like in.xxx
> program for these tests,

fwiw, 'ipvsd' was submitted by me because 'fnord'[1] requires a daemontools
like inetd. And 'fnord' is heavily (and probably explicitly) advised to be
linked against dietlibc.

> unless you want to argue that all programs executed by invsd are going
> to be static linked too, in which case you can start to vastly expand
> you're auditing of dietlibc to all functionality such a in.xxx program
> could use, so basicly to all functionality I would like to see results
> for resD.diet and resD.ds and compare those to resD.glibc

No; this subthread exists to decide whether dietlibc increases performance
of 'ipvsd' significantly. Therefore, only the code executed within 'tcpsvd'
should be measured.

It is obviously that a slow program would increase absolute numbers and
reduce the relative ones. Dynamic linking against glibc would make small
programs significantly slower so I used the fastest possible 'r'.

Ok, telling the 'resD.glibc.*' results at this place was wrong and
related to other statements in this thread. So, please ignore them...

> About your resident sizes results this doesn't take the shared part of
> this into the picture. Please substract the shared part (minus the
> stripped binary size) from these figures since the few basic glibc
> functions used will be resident already in any realistic Fedora test
> scenario.

I am not sure whether this it what you want but /proc/*/status tells:

Name:     tcpsvd.ds     tcpsvd.diet    tcpsvd.glibc
VmPeak:      156 kB          340 kB         1768 kB 
VmSize:      136 kB          320 kB         1636 kB 
VmLck:         0 kB            0 kB            0 kB 
VmHWM:        60 kB          176 kB          504 kB 
VmRSS:        44 kB          156 kB          504 kB 
VmData:       12 kB           36 kB          160 kB 
VmStk:        88 kB           84 kB           88 kB 
VmExe:        32 kB           36 kB           48 kB 
VmLib:         0 kB          152 kB         1304 kB 
VmPTE:         8 kB           12 kB           20 kB 


[1]  http://www.fefe.de/fnord/

Attachment: pgpAOrhvSM0aN.pgp
Description: PGP signature

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]