[prev in list] [next in list] [prev in thread] [next in thread] 

List:       opensolaris-discuss
Subject:    Re: [osol-discuss] OpenSolaris cancelled,
From:       Octave Orgeron <unixconsole () yahoo ! com>
Date:       2010-09-08 19:38:47
Message-ID: 63751.66686.qm () web30808 ! mail ! mud ! yahoo ! com
[Download RAW message or body]

FYI, 256 sockets is nothing. For years universities and governments have linked 
E15k/E25ks together into single image servers, and you can put 72 CPU's in each 
of those servers. Same thing for the Fujitsu PrimePower 1500/2500. SPARC and 
Solaris have had this scaling capability for ages. The issue is costs. This is 
no different from Intel, while they could scale up, the problem is lack of 
chipsets and hardware vendors to support it. Right now, with everyone scaling 
CPU cores and threads, chasing after sockets makes less sense. Which is why both 
Sun Oracle and IBM only have 64 socket servers around, not enough demand for 
something bigger. And it's those 64 socket servers that blow the competition 
away on OLTP, Java, SAP, etc benchmarks, not x86.

Who's going to build the 256 socket Intel Nehalem-EX beast? Unisys?? There have 
been options in the market to link x86 boxes together using infiniband for a 
while, but the market penetration is pretty weak. The big problem for x86 is 
that Intel has upped the bandwidth for CPU->memory, but hasn't done much to 
address I/O. Hell it wasn't until recently that Intel and AMD added IOMMU's into 
their CPUs.. something that has been around since the 90's for SPARC. Until an 
x86 manufacturer makes a truly balanced (CPU/MEM/IO) and RAS capable server, 
there will continue to be a market for Sun Oracle and IBM to petal their 
big-iron. Come on, you still can't hardware partition an x86 box or get past all 
the old x86isms. 


Intel and AMD use RISC cores with CISC converters anyways:) RISC already won the 
battle, it's just that OS's like MS windows and Linux still need the CISC x86 
support in hardware. There was even a time when AMD had partnered with 
DEC/Compaq to make socket compatible systems to swap in Alpha chips. That was 
probably the best CPU to blow away all the others, that roadmap had 8 cores over 
10 years ago! But we all know how that turned out with HP and Intel. I bet if 
Intel and AMD were to dump the CISC conversions and go pure RISC, x86 would 
really scream. But lots of other things to address before they could ever do 
that.

And for all those HPC x86 linux fans, here's something you probably didn't know. 
Back when governments, labs, and universities started to switch over to x86 
Linux, it had nothing to do with performance. It was all about "up-front" costs. 
To replace what a rack of dual or quad socket DEC Alpha servers could do, you'd 
need several racks of x86 servers. Sure those x86 servers were cheaper 
"up-front", but were more expensive to run and operate. The overhead for power, 
cooling, cabling, and management far out-weighed the cost of smaller RISC 
solutions. This is no different today in the Enterprise. x86 is cheap, it's 
fast, but you need a whole boat load more of it to do the same amount of work. I 
can give plenty of examples where x86/Linux requires more physical servers than 
to do the same workload on Solaris with T-series servers. It's not always about 
who can run at the fastest clock speed.

So either you spend the money on the RISC solution which is higher "up-front" 
costs, but easier to maintain and manage due to lower TCO.. or you go with 
Intel/AMD and the Penguin to save "up-front" costs and spend more on your TCO. 
That's the real situation right now in the enterprise. No different than it was 
10 years ago in HPC environments.

 *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
Octave J. Orgeron
Solaris Virtualization Architect and Consultant
Web: http://unixconsole.blogspot.com
E-Mail: unixconsole@yahoo.com
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*



----- Original Message ----
From: Orvar Korvar <knatte_fnatte_tjatte@yahoo.com>
To: opensolaris-discuss@opensolaris.org
Sent: Wed, September 8, 2010 12:17:25 PM
Subject: Re: [osol-discuss] OpenSolaris cancelled, to be replaced with Solaris 
11 Express

How is that nonsense?

IBM has said OFFICIALLY they are going to kill AIX. When IBM say so, is it 
nonsense of me to repeat what IBM said? How is it nonsense? In fact, it makes 
very sense to repeat what IBM say. Actually, it is nonsense to reject IBMs own 
official statements. In other words, it is you that talk nonsense.

When Linus Torvalds said that "Linux is bloated" and I repeat that - am I 
talking nonsense? 




Regarding POWER. AIX runs on POWER, yes. What happens to AIX if POWER dies? Then 
AIX probably dies too. IBM is not interested in porting AIX to x86. Why would 
they? Then they can not charge outrageously high prices for AIX. When people can 
compare AIX prices on x86, to ordinary x86 vendors - they will be shocked.

And yes, POWER has a bleak future. It will die eventually. x86 is today almost 
as fast as POWER7, for a fraction of the price. The biggest POWER7 machine has 
64 sockets. Intel Nehalem-EX scales to 256 sockets. Earlier, POWER and SPARC 
crushed x86. Nowadays, x86 is getting so much resources it has the fastest 
development pace. It will very soon beat POWER (Sandybridge and Bulldozer is 
soon out, and next next generations are soon out long before POWER). 


x86 and POWER: they are competing in the same territory: few heavy weight 
threads. x86 will win this race, there is too much money in x86.

x86 and SPARC: x86 has won in terms of performance

x86 and Niagara: Niagara has found it's niche and wins in multi threaded work 
loads. Niagara T3 and T4 will rock the boat and no other cpu can match Niagara 
in its niche.

Ergo, POWER is dying. AIX will die too. Then it makes sense for IBM to shift to 
Linux. IBM just tries to milk the cow as long as possible. When x86 outperforms 
POWER (or gives equal performance) for a fraction of the price, there are no 
reasons to stuck with ultra expensive slower AIX gear. POWER will go the same 
way as Itanium goes. Itanium is soon dead too - too slow and too expensive. 
POWER will soon be the next Itanium. And of course, SPARC will also die. But 
Niagara will give far more performance than any other cpu, so it might live.

Only way POWER and SPARC can live, is if they give much higher performance (not 
likely as x86 is catching up) or if they are cheaper than x86 (not likely, 
because of the mass volumes of Intel and AMD).

Solaris runs on x86, so it will live. AIX will die. As a coincidence, IBM has 
officially said that "AIX will be killed in favour of Linux on x86". Make your 
bets.
-- 
This message posted from opensolaris.org
_______________________________________________
opensolaris-discuss mailing list
opensolaris-discuss@opensolaris.org



      
_______________________________________________
opensolaris-discuss mailing list
opensolaris-discuss@opensolaris.org
[prev in list] [next in list] [prev in thread] [next in thread]