On Monday 10 October 2005 13:01, S Destika wrote:
> <b> If you use an email client - reply to the mailing list only. Do NOT
> reply to me as it will bounce.</b> <br>
> <br>

Thanks for putting this disclaimer.

> What makes you think on all other counts Linux will fail to scale
> miserably? I would be curious to hear if you pointed out the flaws or
> pointed me to real world experiences where they learnt Linux 2.6 didn't
> scale.

Same goes for Linux. Can you point me to real world experience where Linux is 
used in the 512CPU system you point out? Better yet, can you show me real 
world cases (not issolated) where Linux is being used for 16-way to 64-way 
systems on a daily basis? Numbers is tests don't have much meaning, it's real 
world production servers that do count.

> IOW, show me the benchmarks which prove that Linux has "thoroughly poor"
> SMP scalability? (I am only interested in 2.6 kernel).

This seems that you're admitting that the 2.4 and older kernels didn't do 
well. Isn't the 2.4 kernel being distributed on RHEL? That is what the 
majority of folks are running in production that are using Linux.

> Solaris may edge out in some cases but Linux should be close enough.

I find this comment most intriguing. Previously you stated how much slower 
Solaris was, per SPECweb, which was made out to be unacceptable. In this case 
you show that Linux is "close enough".

The one comment I would like to make is that any given test is not a reason to 
use a server, and any test is not a reason to prove that what company X needs 
to accomplish is the same that company Y needs to also. SPECweb is good for 
showing load on web server. But most companies need more than a web server.

> All of the SPEC benchmarks and my personal experience indicates the same.

Can you tell us a bit more what environments you've run Linux in and Solaris 
in where you've seen the SPECweb results to be true across the board?

> Tell me where exactly Linux fails miserably on SMP - scalability wise.

I've seen trap screens while running SMP kernels on both 2.2 and 2.4 kernels. 
I realize now that you're only interested in 2.6 kernels. At VA Linux 
Systems, who was arguably one of the larger Linux companies in it's time 
(though now dated, I admit that), the servers would roll over from time to 
time and need a reboot. There were cases where they would get trap screens 
also. This was an environment where they ran a web server, mysql database, 
and lots of IP traffic on 2-way systems. This was real world, not 
benchmarking.

I worked a consulting job before coming to Sun where the client insisted on 
running Oracle on RHEL. It was really difficult to get installed, unlike 
Solaris where you do the install and get going. The Linux server required a 
massive amount of configuration, with 3 engineers working for 2 solid days to 
get things stable.

> In my book, "Robbing Paul to Pay Peter" is not good scalability - which is
> what Solaris does. Slow fork, Slow thread creation, slow IPC and till
> Solaris 10 the IP stack sucked badly.

Granted, Solaris is not perfect, and you're correct that the IP stack was 
becoming dated prior to S10. However, that has been fixed and Solaris now has 
a very nice stack in that regard. You're also correct that Solaris is slower 
in some aspect, than Linux. But this is only in specific areas. Overall I 
think Solaris stands on it's own and doesn't need any validation. It is used 
widespread by many companies (yes, many large names).

Worth noting that Linux shops were using the 2.2 kernel at the time Solaris 9 
was released, and the 2.4 kernel was considered to be unstable. I think it's 
ok to compare Solaris 9 if you prefer, but in that case let's compare the 
Linux 2.2 kernel. Sure, things change and improve, that's the case for 
Solaris also.

> Still Solaris proponents were beating
> their scalability drum since Solaris 2.6 - just because opponents were in
> their infancy as far as SMP scalability goes. Same is not the case today.
> Linux 2.6 is fast and reasonably scalable where it competes.

There are hundreds of real world cases where SMP plays a big part of 
production environments and Solaris has been running in those for a long 
time. Who would you use as example of shops running large SMP installs of 
Linux today? I can give ebay as an example of a shop running Solaris SMP, 
arguably one of the largest installs and high traffic servers on the 
internet. Several Wall Street companies have used Linux and moved back to 
Solaris in recent times (I won't mention any names, so you don't have to 
believe me if you don't like). Maybe someone at Sun can point to some public 
announcements that would indicate such.

-- 

Alan DuBoff - Sun Microsystems
Solaris x86 Engineering - Sun on Sun is the way of the future!


_______________________________________________
opensolaris-discuss mailing list
[email protected]

Reply via email to