Re: Re: [gentoo-dev] Creating a dedicated gentoo profile for commercial mysql support

Re: Re: [gentoo-dev] Creating a dedicated gentoo profile for commercial mysql support

[Matthew Marlowe]

>> > 1) Mysql binaries are compiled with icc which could provide significant performance
>> gains for the targeted user base.
>> > Untill icc is fully compatible with portage, I'd rather avoid installing icc myself on
>> customer boxes.
>> 
>> I am really not a fan of this. What is the problem with icc and portage
>> right now? Set CC=icc and compile away.
>> 

See several of the comments in bug #26757.   

>> Perhaps this situation could be fixed rather than adding a hack to work
>> around it.
>> 

I'd hope so, but there is still the issue of mysql being more comfortable with their 
binaries than compiling - but I'd be all for using source ebuilds if we could get away 
from it.  I don't think thats going to happen right away.

MattM

-- 
gentoo-dev@gentoo.org mailing list

Re: Re: [gentoo-dev] Creating a dedicated gentoo profile for commercial mysql support

[Matthew Marlowe]

>> Matthew Marlowe wrote:
>> <too big snip because I'm writing in vncviever>
>> - there are tests that show mysql compiled with icc much faster,
>> obviously only on intel box
>> - you can find a very basic ebuild for mysql-bin at b.g.o/83424
>> - packages for mysql-4.1 have been dowloaded hundreds times starting
>> 2004-12 from b.g.o/83011 and http://www.francesco-riosa.com/gentoo/
>>   very few reported bugs, mostly ebuild related (not mysql fault)
>> - the story is totally different if you look at mysql-5.0 or gcc-4.0
>> currently there are 74 Mb of compile/test log that can document it.
>> - on the server maillist the idea of a slow update / hyper stable branch
>> has been not very well accepted as far as I can remember.
>> 

If we were just talking about a single mysql 4.1 server sure, the gentoo
ebuilds have been great.  I haven't had any problem with it and the mysql
ebuild writers have been very good.

Unfortunately, my clients would like to use mysql 5 (which will released from
beta to production in June/July according to the conference), clusters (which
can get complicated), best possible performance, and commercial support.  

That seems to point towards getting a stable profile, icc, and binaries at the moment.

As far as the stable branch, I've heard alot of devs really like the idea - and several
want to contribute to it, but there just wasnt enough momentum and there were
technical issues involved (such as keeping copies of distfiles around for long
periods).

>> LIFO proposal
>> 
>> - Learn from TV, update your server in "differita" <- translate as
>> needed.  Follow the updates of the main stable branch ... delayed 4 weeks.
>>  You will have a 100 times more tested tree than the hyper stable one.
>> We need a handly way to do this
>> - Apart from the icc compiled ones avoid the use of precompiled binary.
>>   MySQL compiling and running depends from few other things:
>>   = linux-headers  / kernel ?
>>   = glibc
>>   = gcc / compiler
>>   = crypt
>>   = ssh
>>   = readline
>>   = perl
>>   = nsl
>>   = tcp wrappers
>> Ask to MySQL ab what version of theese they certify and build in loco
>> the stuff should be non plus ultra
>> 

Thanks for the list above.  I'll definitly pay attention to all these items.

>> Best regards
>> Francesco Riosa
>> 

Regards,
Matt

>> --
>> gentoo-dev@gentoo.org mailing list
>> 
>> 
>> 


-- 
gentoo-dev@gentoo.org mailing list