Hi,

we are currently looking into buying new hardware, and the decision has
come down to either go for
 - SunFire v490, 2x1.35GHz US IV+
or
 - Fujitsu Siemens Primepower 250, 2x1.87GHz

All other specs are almost the same. I've searched this group for
experieneces with Primepower servers, but everything I got was a little
old. Does anybody have recent experiences with these servers? Sun's
equipment is twice as expensive as the Siemens servers, is there a
catch?

We are going to use them primarily for an Oracle database (9i), on
Solaris 9.

Thorsten

Thorsten Jens wrote:
> Hi,
> 
> we are currently looking into buying new hardware, and the decision has
> come down to either go for
>  - SunFire v490, 2x1.35GHz US IV+
> or
>  - Fujitsu Siemens Primepower 250, 2x1.87GHz
> 
> All other specs are almost the same. I've searched this group for
> experieneces with Primepower servers, but everything I got was a little
> old. Does anybody have recent experiences with these servers? Sun's
> equipment is twice as expensive as the Siemens servers, is there a
> catch?
> 
> We are going to use them primarily for an Oracle database (9i), on
> Solaris 9.
> 
> Thorsten
> 


Did you also take a look at the v490 1.5 Ghz ??  In belgium they cost no more
than the 1.35 Ghz version.  We use the V490 / 1.5 Ghz for general multi-user
servers with >1000 processes running.

We have the FSC primepowers for database / Application servers.

The V490 has 2 cores/cpu.  The 250 only one, but much higher performance/cpu.
Oracle licensing might cost a lot more on the V490.  ( licensed per core ?? )
The V490 can still be upgraded to 4 cpu's. ( not so for the PP250 )

We run several PP650 and M800 and have good experience with them but :
	- most primepowers do not have ALOM management yet.
	- you need specific versions of solaris, extra Fujitsu packages  and get patches
	  etc. from Fujitsu instead of SUN.
	  ( -> twice as much work in a mixed sun/fujitsu environment )
  	- Not every Solaris distribution will work on the FSC Primepower.
	  We had to use Solaris-9 9/04 instead of the most recent version !!!
	- fujitsu has no equivalent of sunsolve
	- fujitsu has no osc website ( at least in belgium ) for online
	  service requests.
	- very poor java garbage collection times in larger configurations.
	  ( eg. bea weblogic with 1.5 Gb heap )
	  Sun uses the VIS instruction set in several optimized libs that are not
	  available in the Primepower Sparc implementation.
	- not so easy to find used spare-parts on the internet for primepower.
           ( if you want to 'upgrade' your systems yourself )

Positive :
	- highest performance/cpu  ( > 1350 SPECint2000 / cpu )
	- lower cost
	- very stable hardware

Peter.

No catch... We have a number of Primepowers... 250s, 450s, 650s and a
couple of 850s.. Great boxes and great customer service... The
Primepowers have pretty much displaced all our larger Sun servers we
where using for are bigger Oracle databases... We still have Sun
servers... but they are being used for our smaller apps...

peter wrote:
> Thorsten Jens wrote:
> >
> > we are currently looking into buying new hardware, and the decision has
> > come down to either go for
> >  - SunFire v490, 2x1.35GHz US IV+
> > or
> >  - Fujitsu Siemens Primepower 250, 2x1.87GHz

> The V490 has 2 cores/cpu.  The 250 only one, but much higher performance/cpu.
> Oracle licensing might cost a lot more on the V490.  ( licensed per core ?? )

That's another concern, indeed. Oracle licences by core, not by cpu
count.

> The V490 can still be upgraded to 4 cpu's. ( not so for the PP250 )
>
> We run several PP650 and M800 and have good experience with them but :
> 	- most primepowers do not have ALOM management yet.
> 	- you need specific versions of solaris, extra Fujitsu packages  and get patches
> 	  etc. from Fujitsu instead of SUN.
> 	  ( -> twice as much work in a mixed sun/fujitsu environment )

The non platform-specific Solaris patches from Sun still apply to
Primepower, though, right?

>   	- Not every Solaris distribution will work on the FSC Primepower.
> 	  We had to use Solaris-9 9/04 instead of the most recent version !!!
> 	- fujitsu has no equivalent of sunsolve
> 	- fujitsu has no osc website ( at least in belgium ) for online
> 	  service requests.

We're going to get a phone support contract anyway, so that's no big
deal.

> Positive :
> 	- highest performance/cpu  ( > 1350 SPECint2000 / cpu )
> 	- lower cost
> 	- very stable hardware

Thanks for the input,

Thorsten

Thorsten Jens wrote:
> peter wrote:
>> Thorsten Jens wrote:
>>> we are currently looking into buying new hardware, and the decision has
>>> come down to either go for
>>>  - SunFire v490, 2x1.35GHz US IV+
>>> or
>>>  - Fujitsu Siemens Primepower 250, 2x1.87GHz
> 
>> The V490 has 2 cores/cpu.  The 250 only one, but much higher performance/cpu.
>> Oracle licensing might cost a lot more on the V490.  ( licensed per core ?? )
> 
> That's another concern, indeed. Oracle licences by core, not by cpu
> count.
> 
>> The V490 can still be upgraded to 4 cpu's. ( not so for the PP250 )
>>
>> We run several PP650 and M800 and have good experience with them but :
>> 	- most primepowers do not have ALOM management yet.
>> 	- you need specific versions of solaris, extra Fujitsu packages  and get patches
>> 	  etc. from Fujitsu instead of SUN.
>> 	  ( -> twice as much work in a mixed sun/fujitsu environment )
> 
> The non platform-specific Solaris patches from Sun still apply to
> Primepower, though, right?

Probably, but we were told that we should download ALL our patches from the
FSC website. I prefer the SUN sites and patches but download only from FSC
to prevent support discussions afterwards.

> 
>>   	- Not every Solaris distribution will work on the FSC Primepower.
>> 	  We had to use Solaris-9 9/04 instead of the most recent version !!!
>> 	- fujitsu has no equivalent of sunsolve
>> 	- fujitsu has no osc website ( at least in belgium ) for online
>> 	  service requests.
> 
> We're going to get a phone support contract anyway, so that's no big
> deal.

Phone support was terrible a year ago ( a call center somewhere in Turkey :
took me sometimes 40 minutes to spell serial numbers and contract id's )
but it's much better now.

>> Positive :
>> 	- highest performance/cpu  ( > 1350 SPECint2000 / cpu )
>> 	- lower cost
>> 	- very stable hardware
> 
> Thanks for the input,
> 
> Thorsten
> 

Thorsten Jens wrote:
> peter wrote:
> > Thorsten Jens wrote:
> > > we are currently looking into buying new hardware, and the decision has
> > > come down to either go for
> > >  - SunFire v490, 2x1.35GHz US IV+
> > >  - Fujitsu Siemens Primepower 250, 2x1.87GHz

But not a T2000?

> > The V490 has 2 cores/cpu.  The 250 only one, but much higher performance/cpu.
> > Oracle licensing might cost a lot more on the V490.  ( licensed per core ?? )
> That's another concern, indeed. Oracle licences by core, not by cpu
> count.

Depends. The T2000 8 core Oracle license is a far better deal than the
2 core V490... Completely different formulas. Last time I read in the
Register anyway : >

One problem with the T series is less builtin redundancy maybe..
I mean compared to the other choices.

The performance is pretty adequate on the V490 running Oracle with a
mere 2 CPUs I
have experience with.

Thorsten Jens <thojens@gmx.de> wrote:
> Hi,
> 
> we are currently looking into buying new hardware, and the decision has
> come down to either go for
> - SunFire v490, 2x1.35GHz US IV+
> or
> - Fujitsu Siemens Primepower 250, 2x1.87GHz
> 
> All other specs are almost the same. I've searched this group for
> experieneces with Primepower servers, but everything I got was a little
> old. Does anybody have recent experiences with these servers? Sun's
> equipment is twice as expensive as the Siemens servers, is there a
> catch?
>
> We are going to use them primarily for an Oracle database (9i), on
> Solaris 9.

You are comparing two different machine types here. The V490 is positioned
as midrange (upgrade to up to 4 dual core CPUs), while the PP250 is more of
an entry level server (comparable to the V240). You cannot upgrade the
PP250 any further.

That said:

It depends on your workload. If the expected load average is < 2-3 then
the PP250 will most likely be faster. For higher loads you should probably
go with the V490 (or PP450)

Regarding notices from other users:

- The SPARC64V do support VIS:
% isalist
sparcv9+vis2 sparcv9+vis sparcv9 sparcv8plus+vis sparcv8plus sparcv8 sparcv8-fsmuld sparcv7 sparc
% prtdiag
System Configuration:  Sun Microsystems  sun4us Fujitsu Siemens
PRIMEPOWER450 4x SPARC64 V
System clock frequency: 220 MHz
Memory size: 4096 Megabytes

========================= CPUs =========================

                    Run   Ecache   CPU    CPU
Brd  CPU   Module   MHz     MB    Impl.   Mask
---  ---  -------  -----  ------  ------  ----
 0     0     0      1100     1.0   5        0.8
 0     1     1      1100     1.0   5        0.8
 0     2     2      1100     1.0   5        0.8
 0     3     3      1100     1.0   5        0.8

[...]

- Service:
  In no way comparable to Sun. No OSC - and the overall support case isn't
  handled as professional as with Sun (origin: Luxembourg/Belgium; in Germany
  support handling was much better, but still not on par with Sun)

- You *can* use patches directly from SunSolve, although officially not
  recommended nor "supported" by FSC, the patches are the same - they just
  arrive at the Fujitsu Patch-Server later (and may sometimes just skip a
  revision).
  Specific patches for FSC add-on cards must be downloaded from Fujitsu
  though.

- Normally every new Solaris Release (and Update Release) should also work
  on FSC platform. The Solaris 9 09/05 (?) release was an exception - it wasn't
  a normal Update Release though (no new features, just added support for
  new Sun hardware)

- If possible stay away from FSC add-on cards. E.g. FSC has its own qfe-Adapter
  (called fjqe). They aren't part of the Solaris CDs/DVDs and have to be
  installed afterwards separately.
  If you want to boot from this cards (e.g. during Jumpstart installation)
  you have to patch the miniroot of the install server and adjust your
  install_end scripts.

- Managebility
  The PrimePower machines have a ALOM equivalent. It is called LAN-console
  and is roughly comparable to ALOM.
  To configure LAN-console you have to install additional packages (from the
  so called ESF (Enhanced Support Falicities)) though.


Our machines have been extremely stable over the last few years. Not a single
hardware defect yet. (Single thread) performance is ~30% faster at same
clock speed than UltraSPARC-III. In contrast to US-III a program compiled
as 64 bit runs at roughly the same speed than the same program compiled as
64 bit. On US-III I have seen performance drops when moving from 32 bit to
64 bit from 5-30%[1].

[1] programs which don't need large address space but could benefit from
    native 64bit "long"s (resp. "long long"s)

-- 
Daniel

Daniel Rock <abuse@deadcafe.de> wrote:
....
> clock speed than UltraSPARC-III. In contrast to US-III a program compiled
> as 64 bit runs at roughly the same speed than the same program compiled as
> 64 bit. On US-III I have seen performance drops when moving from 32 bit to
> 64 bit from 5-30%[1].
> 
> [1] programs which don't need large address space but could benefit from
>     native 64bit "long"s (resp. "long long"s)

Empirically speaking, 
code that's heavy with 64bit integer computation compiled with -xarch=v9 
(64bit ABI) doesn't run any faster than the one compiled with -xarch=v8plus 
(32bit ABI with 64 bit instructions),
because with v8plus, compiler already uses *all* 64bit instructions available.

The only difference between v9 and v8plus are:
1) the address space is limited to 32bit
2) there are less number of 64bit integer registers (32 vs 16) available..

1) helps performance on 32bit 
because pointers are all 4-bytes, 
hence all datastructure sizes are smaller,
so less space in the cache and thus more stuff can fit in the cache.
This effect is a bit more pronounced on US3 than on SPARC64
because SPARC64 are out-of-order (vs US3 being in-order issue)
and generally has bigger caches.

2) hardly makes difference in most integer code
because they tend not to require that many registers 
and even when they do, they run out of registers 
because of pointers and offsets, not because of 64 bit integers).

Overall, on SPARC, there's very few compelling reasons to use 64bit.
Beside 64bit address space, the only other thing 
I can think of is the user trap handling (man __sparc_utrap_install).
-- 
#pragma ident "Seongbae Park, compiler, http://blogs.sun.com/seongbae/"

Daniel Rock wrote:
> Thorsten Jens <thojens@gmx.de> wrote:
> > Hi,
> >
> > we are currently looking into buying new hardware, and the decision has
> > come down to either go for
> > - SunFire v490, 2x1.35GHz US IV+
> > or
> > - Fujitsu Siemens Primepower 250, 2x1.87GHz
> >
> > All other specs are almost the same. I've searched this group for
> > experieneces with Primepower servers, but everything I got was a little
> > old. Does anybody have recent experiences with these servers? Sun's
> > equipment is twice as expensive as the Siemens servers, is there a
> > catch?
> >
> > We are going to use them primarily for an Oracle database (9i), on
> > Solaris 9.
>
> You are comparing two different machine types here. The V490 is positioned
> as midrange (upgrade to up to 4 dual core CPUs), while the PP250 is more of
> an entry level server (comparable to the V240). You cannot upgrade the
> PP250 any further.

Those are the machines Sun and FSC recommended for the job. We're
currently running SunFire 3800, also with just 2 CPUs.

> It depends on your workload. If the expected load average is < 2-3 then
> the PP250 will most likely be faster. For higher loads you should probably
> go with the V490 (or PP450)

Workload is usually around ~1-2 at the moment during normal work hours
(on the older hardware).

[...]

Thanks for all the information!

Thorsten

Seongbae Park <Seongbae.Park@Sun.COM> writes:

>The only difference between v9 and v8plus are:
>1) the address space is limited to 32bit
>2) there are less number of 64bit integer registers (32 vs 16) available..

Floating point; the number of integer registers is, of course the same.
(But that's just a typo, I'm sure)

Casper
-- 
Expressed in this posting are my opinions.  They are in no way related
to opinions held by my employer, Sun Microsystems.
Statements on Sun products included here are not gospel and may
be fiction rather than truth.

Casper H.S Dik <Casper.Dik@Sun.COM> wrote:
> Seongbae Park <Seongbae.Park@Sun.COM> writes:
> 
>>The only difference between v9 and v8plus are:
>>1) the address space is limited to 32bit
>>2) there are less number of 64bit integer registers (32 vs 16) available..
> 
> Floating point; the number of integer registers is, of course the same.
> (But that's just a typo, I'm sure)
> 
> Casper

Not typo :)
v8plus (NOT v8) code can use the same number of floating point registers as v9.
The only difference in register usage between v8plus and v9 is 
that in v8plus only %o and %g registers can hold 64bit, 
and %i and %l are only 32bit.

This is because v8plus is still ABI-compatible with v8
and the register window save area on the stack in v8 ABI
is only big enough to save 32bit portions of %i and %l registers,
making the upper 32bit of %i and %l registers not saved 
through context switch or register window spill/fill.
%o and %g are saved by kernel so no problem using them as 64bit registers. 

As I said in my earlier post, this difference is not big enough
to be noticeable in most applications. 
-- 
#pragma ident "Seongbae Park, compiler, http://blogs.sun.com/seongbae/"

Thorsten Jens wrote:
> peter wrote:
> > Thorsten Jens wrote:
> > >
> > > we are currently looking into buying new hardware, and the decision has
> > > come down to either go for
> > >  - SunFire v490, 2x1.35GHz US IV+
> > > or
> > >  - Fujitsu Siemens Primepower 250, 2x1.87GHz
>
> > The V490 has 2 cores/cpu.  The 250 only one, but much higher performance/cpu.
> > Oracle licensing might cost a lot more on the V490.  ( licensed per core ?? )
>
> That's another concern, indeed. Oracle licences by core, not by cpu
> count.

There is a promotion going on right now at Sun that gives you Oracle
for free
upon purchase of an US IV or IV+ server, which should apply to the
V490.
Don't know the details about the promotion as regards per core
licensing,
you might want to take a look at:
http://www.sun.com/emrkt/db4free/index.html?cid=96412

For most database applications I would expect the V490 to provide
better
throughput since you are running 4 threads in parallel vs. 2 for the
250 even though
the performance per core is better for the PrimePower.  Also, as
mentioned above,
the 490 is easily upgradable to twice the throughput if needed in the
future.

/Karl D

gerryt@ wrote:
>
> Depends. The T2000 8 core Oracle license is a far better deal than the
> 2 core V490... Completely different formulas. Last time I read in the
> Register anyway : >
>
> One problem with the T series is less builtin redundancy maybe..
> I mean compared to the other choices

That's going to be the problem with the T series. We need two Fibre
Channel links, mirrored boot disks, etc.

Thorsten

Thorsten Jens wrote:
> gerryt@ wrote:
> > Depends. The T2000 8 core Oracle license is a far better deal than the
> > 2 core V490... Completely different formulas. Last time I read in the
> > Register anyway : >
> > One problem with the T series is less builtin redundancy maybe..
> > I mean compared to the other choices

Actually I posted w/o looking at the details at the Sun site...
More redundancy (in the form of hot swapping anyway) than I recalled
sorry.

> That's going to be the problem with the T series. We need two Fibre
> Channel links, mirrored boot disks, etc.

Arent 4 SAS disks enough for a mirrored root disk? : >
Looks like there are 5 PCI slots too. Option # SG-XPCIE1FC-QF4
is a 4 Gb FC host bus adapter.

Seems youll have to come up with better excuses :  >

gerryt@ wrote:
> Thorsten Jens wrote:
> > gerryt@ wrote:
> > > Depends. The T2000 8 core Oracle license is a far better deal than the
> > > 2 core V490... Completely different formulas. Last time I read in the
> > > Register anyway : >
> > > One problem with the T series is less builtin redundancy maybe..
> > > I mean compared to the other choices
>
> Actually I posted w/o looking at the details at the Sun site...
> More redundancy (in the form of hot swapping anyway) than I recalled
> sorry.

And I was looking at the T1000, thinking they only differed in CPU
configurations. I'll have Sun make me an offer fot the T2000, to see if
it's worth considering.

Thorsten