SUMMARY:RealWorld Comparison: Solaris 8 - 9 - 10 on e450 calibre hardware, Baseline Performance "gut reaction" ?

From: Tim Chipman <chipman_at_ecopiabio.com>
Date: Fri Mar 04 2005 - 08:51:47 EST
MANY thanks to all those who replied ; including (no particular order):

Tom Crummey    Peter Ondruska  Rich Teer       Martin.Wheatley
Tristan Ball   Michael Horton  Octave Oregon

The general consensus has been very consistent, and very positive. Many 
reports suggest that observed performance is definitely "increased" ; 
this feeling being especially the case with folks who have been testing 
solaris10 for a while now. "Worst case" comments received suggest that 
there will be minimal or no boost (but no loss of performance). In light 
of these comments, Solaris10 seems the preferred platform of choice, 
where choice exists.

In light of this, I'm proceeding with testing of solaris10 on my 
development e450 system. Alas I have no significant progress to report 
yet due to hardware issues (infamous SRC/P raid controller being 
desupported, and also the A1000 disk array it seems) that I must 
workaround before proceeding in a meaningful way. [the clean installs 
did go smoothly, and the new installer GUI is rather pretty though :-) ]

Again, many thanks to all respondents. To be thorough, I've included 
comments below (Again in no particular order).

---Tim Chipman


===paste===


I've been testing Solaris 10 since build 63. I've tested it on a wide
variety of hardware (Ultra60, Netra X1, Ultra10, Ultra2, no-name PC's,
and even a compaq laptop). My overall opinion is that Solaris 10 is
faster than 9 or 8. There are some good reasons for this:

1. Better memory and resource management built-in. 
2. SMF brings up services in parallel, instead of the old sequential
init boot process. Faster booting is good:)
3. Better compilers for Solaris. Since Sun bought Forte a couple of
years back, the compiler optimazations have made huge improvements in
run-time binaries and libraries.
4. Better schedualers (Fair-share, Real-time, Time-share, etc).
5. Better threading models and API's. The enhancements are transparent
and require no recompile to take advantage of. New applications can
take advantage of them to a greater extent obviously;)
6. UFS performance has improved dramaticly since Solaris 8. In some
studies compairing VxFS and UFS /w logging, UFS performed better or
equal.

Obviously there are tons of new features in Solaris 10 that make life
easier for sysadmins, users, developers, etc. Definitely take a look
the "What's New" document on docs.sun.com for Solaris 10. 

At home, I use an Ultra60 as my primary computer for everything.
Solaris 10 is a lot faster than 8 or 9 was on the same hardware. When I
say faster, I mean that it's faster booting up, responsive in JDS/CDE,
and responsive for doing work. It's always nice to see new versions of
Solaris work on my old Ultra60 and increase performance. 
------------------
we've upgraded about one year ago couple of e250 from Solaris 7 to 9. I 
would say performance is about the same or slightly better; running 
Oracle 8.1.7 and 9.2.0 databases.

------------------
We run hardware from Ultra 5 through to V240 all on Solaris 9 now (some
started on Solaris 7). We've not noticed any degradation in performance
between 8 and 9 and given our hardware replacement cycle it doesn't make
sense to keep multiple versions of the OS, so we tend to upgrade the OS
every year or so. We're currently evaluating Solaris 10 on an Ultra 10,
but we're not far enough along to say whether it's faster.

------------------
We've been running SOlaris 10 as part of the Betat programme for the last year
of Ultra 60s, ultra5s, blade 100s and 150s, 1000s, e250s and E220s and the general feeling from
the users is that the machines perform better than they did running Solaris 7 or 8 (we never ran 9 on
any user platforms)

------------------

Depends on your application I'm afraid.

Most user level applications, or mainly CPU intensive applications won't
see much of a difference.

There may be more of an improvement to IO bound applications,
particularly those capable of using AIO.
My understanding is that networking performance is considerably improved
on 10, in particular in regards to CPU utilisation, which is likely to
be a throughput win for Gig-Eth on slower hardware, which tends to be
CPU bound

I can't give you firm benchmarks on either however.

Realisticly however, have a bit of a think about how much time your
application spends waiting on the OS - generally it's going to be in the
5-10% range, so even a 200% OS speed improvement is only going to be
2.5-5% to your users. Disk IO can be the exception, because of the
absolutely huge wait times involved. Networking again, _can_ be an
exception, as there are a number of network services which require
little application processing, but lots of data or packets on and off
the wire - which means the OS is a bigger slice of the pie.

------------------
does this really meet the requirements for the sunmanagers' mailing list?
pleae read the sunmanagers' faq!

------------------
WHen I upgraded an Ultra 1 web server (256 MB of RAM, 170 MHz CPU),
the difference was so noticable that customers asked me what I'd
done to improve the performance!

I expect no less from other HW, and even more improvements with S10
on supported HW.


>> -is the baseline performance for Solaris 8 / 9 / 10 .. on this calibre
>> hardware ... to be expected to be similar ..or.. somewhat slower as we
>> get "more current" .. or more-or-less unchanged? [ie, increased overhead
>
Given enough RAM, newer versoins of Solaris tend to have BETTER 
performance than older ones. Methinks you';re too used to M$'s constant 
resource hogs!

>> -has anyone installed solaris9 or 10 to a solaris8 box, and observed
>> "perceived" (or benchmarkable/quantifiable) increase to performance ..
>
Yep; see above.



=====original posting text follows=====

A rather classic/typical/lame question, but as always somewhat difficult 
to get "real world from the trenchs" feedback / observations.


Now that Solaris 10 is truly released, Solaris9 is well and truly ... 
old enough for stuck-in-the-mud types to consider it   ... I am 
curious if anyone with first-hand experience has comments on behavior of 
Solaris9 and 10 on older hardware (e450/e250/e3500, 400mhz CPUs / 2gigs 
ram in the systems) that previously ran Solaris8.

ie,

-is the baseline performance for Solaris 8 / 9 / 10 .. on this calibre 
hardware ... to be expected to be similar ..or.. somewhat slower as we 
get "more current" .. or more-or-less unchanged? [ie, increased overhead 
of the OS itself on the system resources? vs  "increased efficiency of 
more elegant code" to streamline performance? ..etc..]

-has anyone installed solaris9 or 10 to a solaris8 box, and observed 
"perceived" (or benchmarkable/quantifiable) increase to performance .. 
(gasp?!)

Certainly I'm aware there are a number of specific features that become 
available as the more-current OS is selected (zfs, dtrace,etc) but such 
details aside, I was just wondering in general if people are doing this 
sort of OS upgrade on legacy hardware ... or if the more typical 
"migration path" people choose these days is to buy newer hardware 
running the newest OS, migrate apps and then retire the legacy hardware.

I'm certainly going to muck around with a few such 
migration/installations on some test systems here for a lark, but I 
always appreciate getting some context / info from others who may have 
done similar work (and such things, I find, are never ... clearly 
available ... from Sun and are often tricky to track down with Google even).

Comments / thoughts are certainly very welcome. 

Thanks,

--Tim Chipman
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers
Received on Fri Mar 4 08:52:21 2005

This archive was generated by hypermail 2.1.8 : Thu Mar 03 2016 - 06:43:44 EST