[dundee] How do you do yours?

[Andrew Clayton]

On Fri, 2003-04-04 at 21:16, Mark Harrigan wrote:
> On Fri, Apr 04, 2003 at 09:03:07PM +0000, Wm. G. Urquhart wrote:
> > On 4 Apr 2003, Andrew Clayton wrote:
> > 
> > > 
> > > I presume, dual boot with Slowaris....
> > 
> > Hi Andrew,
> > 
> > Is there anything I should know here, as I don't think the above is a 
> > typeo. The Sun hardware I have is very, very respectable in the 
> > performance dept.
> >  

Absolutely, there is nothing wrong with the Sun hardware....


> > -- 
> > William.
> 
> I think it's just a little bit of the old Fear, Uncertainty and Doubt. :)
> Some people do regard Solaris as slow, it's often down to the
> comparitive performance of Sun processors vs competitiors which is frequently
> less, however the reality of the situation is usually much different as the
> Sun systems benefit from good internal system bandwidth and excellent 
> scalibility.
> 

I'm not blasting the UltraSPARC processor/Architecture here.

I hold Sun hardware in quite high regard having worked with it in the
past.


> Intel Linux people often think Solaris is slow because of the poor
> performance of gcc under the environment.
> 

Yes, GCC under Solaris is probably not the best performer, in fact GCC
on anything probably isn't the best performer.


> Essentially I suspect Andrew was merely extracting the urine...
> 

Not really.


> PS Andrew: If you weren't please correct me, but I've never heard of any huge
> problems with the platform that have any basis in fact.
> 

Well it was just a wee joke along the lines of Internet Exploder, MS DOG
and Nutscrape (you can probably work out what these are). I was really
just referring to the fact the Sun have compromised Solaris's
performance on the low end to gain better performance on the high end.
This is a well known fact. It you want specifics, well one is Solaris's
implementation of what would be /dev/poll under Linux, they are fast for
large no's of fd's but slow for small no's of fd's. Also due to the high
level of fine grained locking going on the solaris kernel. it can scale
into 100's cpu's but takes a hit on the smaller machines. 

So unless your UltraSPARC'S have more than say four processors then
Solaris will perform no better (likely worse, depending on what you
measure) than Linux.


For some microbenchmarks of Linux vs Solaris check out
http://www.ultralinux.org/faq.html#q_1_15
of course this is pretty subjective stuff, but makes a nice point I
think.


Hope this clears some issues up.
 

--
Andrew Clayton