Not sure what was driving this article, but the summary is pretty much spot on:

a) The APL boxen (M-series) are filling their market niche well

b) The Niagara (T-series) machines are excellent for the correct workload

c) The x64 machines are a good series which should be selling better than they are.

What the author misses is how the servers and software offerings become more than the sum of the parts. With Solaris 10 containers (zones), it is a reality now that many houses are allocating capacity to projects rather than servers. With logical domains, it is becoming possible to effectively carve up the larger multi-socket niagara machines for general workloads. And the xVM launch a couple of weeks ago put in place the overarching framework in which the SPARC and x64 technologies will be virtualised and the management framework which is to accompany them.

So Sun clearly is going through a major renewal of its product lines (with guaranteed backwards compatibility) and maintaining sales while doing so. The final piece of this renewal will be Rock, but the APL systems stand on their own merits.

If Rock doesn't cut it, Fujitsu has an 8 core Sparc64 chip coming "soon" called Venus. It should be an easy shoe-in to the M-series boxes. Shame that Rock is late and probably not that great. Maybe Mr Erlang (Joe Armstrong) was right when he said that transactional memory was like trying to cut out a cancer (the cancer in question being shared memory contention).

The Achilles heel of the Coolthread series of chips is the very poor single thread performance. It's fine if you have an application that can multi-thread, and where the response time doesn't depend on how fast you execute a single instruction stream. Unfortunately there's a lot of those sort of applications out there that are affected. Things get worse on the Coolthreads processors once you start winding up the throughput - those 8 hardware threads on a Niagara 2 core contend for just two integer and one FP processing unit. As utilisation winds up then the threads in execution slow down further due to this contention so your hardware thread might start looking more like a 700MHz (or slower) processor than a 1.4GHz one. it's not as bad as it might appear, as the core can make use of processing resources when a thread is stalled for something like a memory access, but it can have a very visible effect. it's best to view these hardware threads as virtual CPUs in the same way that guest machines under VMWare can be slowed due to contention.

A single or dual socket Sparc64 won't pirate sales from Coolthreads - it will fill a gaping hole in SUN's line up for SPARC servers; an entry level server with half way decent single thread performance. The Niagara series will always appeal where you need SPARC and want the best throughput per watt (the SPARC 64 is anything but power efficient), but the coolthreads architecture is not the answer to everything.

Nb. this whole things about hardware thread support is something that affects many modern processorrs SPARC64, Itanium and Power all have support for a coarse-threading model (and Intel used to have HyperThreads). They all trade a bit more throughput for much reduced single thread speed at high utilisation levels (and distinctly non-linear relationships between reported CPU utilisation and throughput).

Victoria Falls and the Zambezi river are both on the border of Zimbabwe, the country with the world's higest inflation rate and fastest shrinking economy. Both used to generate a lot of money for the country through tourism, but now nobody wants to go there.

The Botaka moniker gave me a smirk too. Here's to hoping we never get to a Shikaka prototype.

http://en.wikipedia.org/wiki/Ace_Ventura:_When_Nature_Calls

Help...we have fallen and can't get up

Our stock is at a dollar and change (ignoring the ponytail reverse split)

Everyone wants a new CEO

The OEM of Fujitsu gear is not the answer.

The T boxes have only done really well replacing the ancient SPARC install. Any we have been quiet about ROCK for a very good reason.

We just had another RIF day, it was more random than ever.

Andy and John are rarely around...it's like whats left of the top talent may not have left...but they are not around much

Just not at the same time, in fact you get a maximum of 32 threads at one time, the rest are all stalled, or not in use, and in between you still have the too small cache trying to service all those parked threads as the core jumps from one to the other. And to make it to a two-way they had to chop memory controllers off the chip? Ouch - memory latency was already bad, so now you have even worse memory performance than a single chip, and you're still trying to cram, store and retrieve all the data for all those stalled threads. I'm sure T2+ is just great at what it does best, it's just that's not what 99% of business apps do (that's even if you can find the app that's been recompiled to run on Slowaris 10 on Niagara). And the green-savings argument/fallacy would be a lot more compelling if the T2/T2+ didn't cost so much more you could buy a dozen Xeon blades and run them on the difference! In the meantime, Rock is just mroe vapourware to try and hide the age of UltraSPARC. Yes, the SPARC64 boxen are filling the gap, but then Sun doesn't have anything else for the job so they'd better pray Fujitsu can go on making a better SPARC chip than Sun can. They're only other hope is Intel's Xeon - oh how the bigheaded have fallen!

"I'm sure T2+ is just great at what it does best, it's just that's not what 99% of business apps do (that's even if you can find the app that's been recompiled to run on Slowaris 10 on Niagara)"

Last time i looked there is more than 1% of the biz apps out there running JAVA. Since that's why Sun created the T line. I think they know their own software(java) and how to get more out of it.

As for other uses we use a 5220 with Oracle 10G dB for OLTP and it smokes the old E480 in most cases (long running reports being the exception). Given the cost of these servers and you use them wisely JAVA app/web and OLTP you will be very happy.

It seems that you have a strong grasp of the technology and the obvious but sound green on the hands on. Its nice being a critic that has never used the technology. Kinda like reviewing a movie without watching it.

Re: Steven Jones post,

IMO, and judging from all empirical and analytical data on Sun's business, the actual achilles' heel of Sun is their completely myopic focus on Solaris,

i am so sick of the people running Sun, i have no words to describe how stupid it is not to support Red Hat on the T2, it just is the single biggest amateur move in the IT industry...

they have everything they need to deploy top-bottom systems on pretty good middleware, including Glassfish and MySQL, and instead they spend all their money and focus on promoting and sustaining openSolaris...

its a joke of an operation in Santa Clara, everyone in the engineering ranks and the middleware are top-notch, the executive management couldn't run a shoe store (apologies to shoe store managers), which is what they will be looking at when they are finally kicked out...

i am so over Sun executive management...

Maybe Sun needs to stop annoying its older customers. I'm not running a beta operating system (even if its a few years old). If you can't build a core OS that doesn't need essential updates every few weeks, then let me use the older one on your new hardware. Someone at Sun might want to look at the low end of their sparc line and figure out they don't have one anymore. They also ask themselves how people are supposed to put the new bits of gear where the older bits of gear used to be. We bought racks when the x1/v100 was the hot new thing and now their seem to be two models (t1000 and soon to be pulled netra 210) that fit in the same space as a v100 and nothing as lean as a x1. Right now Sun has exactly one machine that meets my requirements and thats the netra 210 and its scheduled to be pulled from the line up.

What, no-one saying "Matt, that's just what you always post, just anti-Sun sh*t"? Maybe it's because they recognise that T2+ still does nothing to fix the problems with the whole concept. It still doesn't scale and has very limited appeal, and ripping out memory controllers to go MP just looks like a performance-sapping kludge. Or maybe all the Sunshiers are busy thinking something along these lines....

Look at the Intel Atom - it's a small core CPU with lowe power requirments that is still butcher than a T2/T2+ core, yet runs standard x86 binaries. Intel delivered it on time and to much aclaim and interest. What if Intel made a multi-core Atom MP for servers - suddenly there's a low-power alternative to T2/T2+ that can run normal x86 Windows and Linux with the advantage of that cheaper Xeon infrastructure. Think of a multi-core Atom MP that can run single-threaded apps or copy Niagara and do the parallel mult-weiner threads thing, only much cheaper and without you having to run Slowaris. Think that's a wild idea? It would be easy for Intel to make a multi-core Atom using Xeon technology. Now, think of a multi-core Atom in a multi-socket blade that fits in standard blade chassis such as HP, IBM and Dell already have in most corporate datacenters, taking advantage of the existing Wintel/Lintel ecosystem - already we have a system that scales beyond T2/T2+, and that's before we start thinking about a socket-compatible Atom MP that could make use of cell-board technology to scale to 128 real multi-core CPUs rather than just 264 part-time threads.

You see, if the weiner-multi-thread school ever does get popular, Intel would swallow up the market with a product that would do more than T2+ at a lower price and with a greater range of ready-made apps, and do it almost overnight. If you wanted to, you could even run Slowaris x86 apps (that's the "free" Slowaris Sun doesn't make any money off) alongside Linux and Windows, or even under VMware of Hyper-V instances - more trusted and widely used than Slowaris containers. Which leaves expensive and scale-constrained T2+ just running expensive and limited-appeal Slowaris. If that ever happens then Sun will stop the Niagara train overnight, and just sell SPARC64 and Xeon and probably Atom MP (unless AMD manages to get Opteron back on track and get their own multi-core low-power Son-of-Geode x86 CPU sorted). Which means Sun's bread-and-butter becomes being just another "me too" x86 vendor.

So, to sum up - Rock is still vapourware, and if it ever sees daylight it will have been made obsolescent by Power and Itanium advances, let alone Xeon/Opteron; Niagara is a dead-end just waiting to be swamped by the x86 tsunami; and the SPARC64 stopgap-that-became-a-main-product-line is completely dependent on Fujitsu's determination to soldeir on with a product being eaten from below by cheap Wintel/Lintel; Sun's Galaxy x86 range just doesn't cut the mustard against smarter x86 rivals; and they're waiting on a court case to see if they will have to pay royalties to NetApp for Slowaris's ZFS being a clone of WAFL; and Sun is so deeply unpopular with the Linux community they would rather buy Dell than Galaxy (!!!). Given that Fujitsu is probably going to be focusing on what will happen with the European and American businesses if the disruptive FSC break-up with Siemens goes through, and suddenly Sun doesn't seem to have anything solid to plan on other than x86 and Windoze.

If you want RHEL support on SPARC, talk to Red Hat. The Linux kernel supports SPARC and Ubuntu and Gentoo are supported by Canonical and the Gentoo community on the T1000/T2000 respectively. Ubuntu is supported on the T5120/T5220 servers. If the more popular distros aren't supported, talk to the OS vendors. I'm sure if there's sufficient demand the OS vendors and Sun could get together and sell their products. If there isn't....then you're out of luck.

When Sun went from 1 chip to 2 chips they only got 60% more performance. I think they are holding back Batoka because they are still trying to jerry rig a benchmark to try to show scalability in an architecture which was designed to only be small sparcII cores on a single chip

I was assigned to be part of the team at Red Hat that told Sun we'd port, but only for $100m per year in cash payments. So we basically told them to shine on. I wouldnt blame Sun, I'd blame the arrogance of the Red Hat team who believed we ruled the universe. It's just amazing how easy it is to blame Sun for everything, at least they're inventing stuff.

Redhat is getting most of their customers from Sun, there is no way they are going to help SPARC survive.

Danny Sugarman

RIP....Sparc is dead....RIF'd to death

Some great discussion points about T2, SPARC, threads etc.

But what about comparing the performance and price of T2 systems vs standard x86 xeon based systems also available from SUN?. I am an Intel employee and wanted to share some fact.. Don't beat me up yet, until you have had a chance to read what I wanted to share below.

If your Operating system of choice is solaris, then you can run solaris on xeon (and SUNs deep knowledge of this OS) or if you want a change of scenery then you can choose linux or windows on xeon (also available from SUN)

I wanted to share with you some publicly available information comparing T2 systems with their xeon equivalents. Based on system capability and price I compare a 1s T2 system ( with 32GB memory this will set you back about $24,000) vs SUNs 2S Xeon ( same memory for about $11,000) and a 2s T2 (64GB memory $56,000) vs SUNs 4S Xeon (about $33,000).

- 2S Xeon outperforms 1S T2 across a range of workloads (never mind the significant price difference). Taking '1' as a baseline, what I have seen on spec.org and sap websites is higher results for xeon. 10% better on SAP, 40% better on specfprate, 60% better on specintrate, 65% better on specjbb

- 4s xeon outperforms 2s t2. 30% better on specfprate, 80% better on specintrate and 40% on specjbb, 23% better on sap

Now this is comparing a mixed bag of operating systems, but even if you take solaris on T2 vs Solaris on Xeon, Xeon outperforms in a solaris environment

- Sun Blade x6250 15% better on specintrate, Sun Xeon 7460 405 better on specjbb, Sun Xeon 7460 10% better on SAP.

So for me as an IT buyer it would be about buying the best performance product from SUN, retaining my Solaris OS and getting all this at a fraction of the cost of a T2 system

Eoin McConnell

I'm amazed that Matt has so much time to rant everytime Sun release a new product - his motives are questionable as they are extreme. On a more postive note, Ubuntu have a port of Linux for the Coolthread T-Series line. See

https://help.ubuntu.com/community/Installation/Sparc

for details if you want a choice of operating system.

I've seen a lot of organizations opt for Red Hat instead of Sun, or in most cases run both side by side. But you are not into significant enterprise computing if you think people actually want to switch systems that really matter to RHL, or any linux for that matter. How can you completely ignore IBM UNIX? I don't think you work in enterprise computing.

But there is a suprising similarity in windows and linux.. they are both great for the small hardware.

Whereas the AC Sunshiner doesn't even try to defend the products. Question my motives all you like, I'm sick and tired of having you Sunshiners sprouting rubbish for years, seems you guys can't take a bit of stick in turn. Your sense of humour seems to have evaporated with the Sun stock price!

I ran a large trial of some Sun T1000's vs some non-SUN Xeon 3Ghz machines (16GB RAM), and a few AMDs. One thing that was clear very soon into the trial (apache web serving, r/o mysql access, php) was that the Intel and AMD kit was significantly faster with their response times (factors of 2-3) in the low end arena where only hundreds of simultaneous requests were made. But the T1000s began to really shine when the workload became intense, leaving the AMD & Intel kit for dead (literally in some of the trials). The x86 performance began to degrade quickly, yet the Sparc's with their magic thread/cores with 32GB RAM began to enjoy the massive load and respond very well, really perking up when the going got tough. We repeated the trial under Linux (originally Solaris10), and the same patterns were seen, although Linux was measurably worse overall (we didn't bother with Windows).

For Gazillions of requests, the Sparc's were measurably better, but they were poor in the low end throughput and response times (probably due to 1.2Ghz clock cycles vs 3Ghz). Those cores come into their own, especially with the RAM cycles being better taken care of than x86 can do.

For live, we installed x86 and the sparc behind load balancers. When the traffic is significant, they do more work than the x86 kit - thanks to the wonders of load balancers working out which responds best.

So, what you are saying is that if Sun would have payed Red Hat $100M per year, they could have Linux on the T2 boxes, and supported by Red Hat with maintenance updates going forward?...and you are implying that Sun thought that was too much to pay?...

thank you for this post, this is the first time i have ever heard of a definitive negotiation between Sun and Red Hat around Linux on SPARC/T2, and will force me to break-down and respond with my own take on this decision by Sun:

bahahahahaha...

they spend $100M on openSolaris sales and marketing in a month and that gets them negative positioning, so their decision to not spend the $100M/annum is further proof that no one at Sun (who can make a decision) knows how to run a cost-benefit analysis, let alone understand the IT marketplace...

this is perhaps the most useful piece of knowledge i have received on Sun's management in some time: they can't spend $100M/year on Red Hat Linux, so they watch their entire business go away, what an absolute joke the Sun executive management team is...

Jonathan, you couldn't manage your way out of an introductory MBA marketing class, let alone a start-up that has no discernible value (Lighthouse), let alone a Fortune 500 company, let alone an IT vendor with the assets to change how systems are implemented worldwide...

in other words, i am ashamed that i even had to indirectly take a top-down order from you, like that Orion garbage you came up with, you have zero idea what you are doing...

I guess once they signed an extension to the Fujitsu OEM they didnt need to throw any more money at the chip

"....But you are not into significant enterprise computing if you think people actually want to switch systems that really matter to RHL...." Really? How about the largest telecoms billing system in Europe, which I believe is also the largest Oracle instance in the UK, which was a Solaris on SPARC deployment wall-to-wall, but is now Red Hat on Superdomes, does that qualify as "enterprise"? Take your head out of the sand and realise there is real sunshine out there as opposed to the Sunshine you've been spoonfed.

"Take your head out of the sand and realise there is real sunshine out there as opposed to the Sunshine you've been spoonfed."

Good point, and that must be the difference then because I see it from a US Federal spoon and you're coming from a UK/Europe telecoms view. For instance, I don't work there, but it would be some pretty wild shit to read that any important system at the Air Inteligence agency switched to Red Hat linux.

I still opine that Redhat's road to Redmond is not through Sunnyvale and they still have IBM to contend with even without Sun.

Interesting comment about 'But you are not into significant enterprise computing if you think people actually want to switch systems that really matter to RHL, or any linux for that matter'. I agree that you are not into enterprise computing until people switch systems...but people are switching systems away from SPARC and over to RHL/Xeon (or chossing to stay on solaris but deploy solaris on xeon)

Telecoms are in fact choosing to move enterprise computing workloads from Solaris/SPARC over to RHL. Check out Telefonica in Spain. they just moved their mission critical online charging system (based on Oracle) over to combination of RHL on Intel Xeon based servers

http://www.intel.com/references/pdfs/Telef_Prepay_FV.pdf.

hope this helps

cheers

Eoin

... is that Sun is bleeding customers like a stuck pig. Period. The only people still buying Sun are the Sunshiners, who would buy a G4 XServe if Sun painted it that weird blue they changed to. I've been dealing with Sun since sun4c, and I'm making like a tree. I'm sure all these Sunshiners would need clean pants if they saw an executive briefing, but I actually make purchasing decisions and have to consider cost to performance benefits.

The only people excited about Rock (which is about to get delayed again) are the same ones who think nobody would ever dare moving away from Solaris. Sure, that's why IBM's Migration Factory is busier than ever. I can't get anything approaching an acceptable level of support on -commercial- Solaris, forget the per-user software products, and I should switch to OpenSolaris or RedHat? How about "no, I'm not that stupid."

That's okay. My VAR is still trying to figure out how to do half the quotes I've requested, and requests for help from Sun continue to go unanswered. They just get referred to the spec sheets on Sun's website.

Hi mis-informers.

Re : Intel vs T2 processors have a look at a quad core intel vs a T2, "The Hare and the Tortoise" : http://blogs.sun.com/mrbenchmark/

Not nice reading for your Intel pride I suspect.

Also I work for a UK telecoms now and we moved away from RedHat, you never counted support costs, you never mentioned support quality either. We ran a benchmark comparison of Sparc T2 vs your current Itanium's in the same price bracket and it wasn't Itanum who won.

Welcome to real world where a support vendor who can actually get your system up and running when issues occur is a genuine asset and where real world performance is what counts.

As an aside, Matt Bryant, who do you really work for? HP? A HP reseller? Your track history is not good, I would really love to know who you work for. I wonder if your simply an alter ego setup by a HP sales droid, geneating FUDD.

www.sunsucks.com, go get Sun benchmarks (m-values) for a 5220 and a 5240. I'd love to see where you got 60% improvement from. The figures I'm looking at show above 90% extra, are you sure your quoting a real figure?

Sun's direction into x64 is clearly an indication of where the market is heading. There are many companies willing to assist with a SPARC to x64 migration to Linux or Solaris x64. Take Transitive. You can use Transitive's QuickTransit product to quickly migrate SPARC applications onto x64 servers immediately saving support.

Also those of you languishing on SPARC T series servers check out what Azul system appliances can do for your app tier.

I am partial to HP ProLiant servers which provide Windows, Linux and Solaris x64 support. HP provides more flexibility and more choice at much lower costs with none of the roadmap risk I see with Sun.

How can I get Sun MValue? Every time I ask for them the basstards throw an NDA at me.

I keep telling them HP has PQRT which they gladly send over and IBM even has their rPerfs on the external website.

I am so sick of Sun and their NDA's. I bet the MValues are just made up bullshiit anyways.

I checked out the www.sunsucks.com site it says something about

SPECjappServer2004: shows when you go from the 1 chip system T5220 to the two chip system T5240 (Maramba) the performance only improves 66%?

Is MValues a more relevant performance metric than Java on a T2?

"....Also I work for a UK telecoms now and we moved away from RedHat...." Really? Strange, I have friends and aquaintances who work on the development and support of BT, Orange, Vodafone, T-Mobile and 3 biling systems here in the UK, and we swap information (and salesgrunt-baiting stories) regularly, and not one has mentioned anything about T2 with any billing system, especially not one using Oracle. Are you talking about some tiny web-based ring-tone provider?

"...., you never counted support costs, you never mentioned support quality either...." Hmmm, consensus seems to be Sun support has sucked for years, and RedHat/Novell/Mickey$haft support is better and cheaper. The rise of Linux and the drop in Sun Slowaris and SPARC sales would seem to support this. But, I suppose there is always an exception to the rule, and you may be one of those rare and lucky individuals.

".... We ran a benchmark comparison of Sparc T2 vs your current Itanium's in the same price bracket and it wasn't Itanum who won....." Damn, I could really use that pointing-and-laughing icon right now! Please, the older generations of Itanium even beats T2 on Java on the many SPEC benchmarks, and T2 would need a massive discount to match the current Integrity range, so I'm pretty comfortable in suggesting your benchmark may have been squewed to provide a Sunshiner result. Still, if your business really runs like your benchmark then T2 would seem to be better, so enjoy. It's just your business seems to be unlike the overwhelming majority of businesses out there.

"....As an aside, Matt Bryant, who do you really work for? HP? A HP reseller?..." Whilst I'm not comfortable revealing whom I work for (sorry, I really don't need any more spam emails, thanks), or even if Matt Bryant is my real name, I can confirm I don't work for HP or an HP reseller. Hypocritical question given your posting as AC, but please feel free to post all your own details and prove your impartiality.

I can't comment on support costs or support qualify as Intel don't sell server systems. Did you run a benchmark comparison of UltraSPARCT2 vs Xeon?. Comparing a 1s or 2s T2 vs Itanium is really comparing Apples and Oranges. Would suggest that you compare a T2 with a its current xeon equivalent in terms of system configuration and look at the performance and price of the systems. Hopefully some of my comments posted earlier might help

hope this helps

Hi Eoin, did you open the link pasted earlier? Sun are showing benchmarks against a Xeon unless I'm much mistaken.

You'll note the CPU speeds in the afore-pasted link are not the most recent but you'll also notice the curve facing downwards when more than about 8 threads are running. Not encouraging for a server that might get hit with a load spike for example, I doubt the curve would change even with higher frequency Xeons unless you also added in better multi-threading support into Xeons?

There are several other peoples posts in other reg articles which stated they benchmarked and the Intel gear showed fantastic responses to begin with then flatlined when the load came along, matching Sun's "theoretical" benchmarks.

Sorry Eoin.

Matt, I don't really care for your baiting, I value my privacy and my respect also so am trying to comment fairly. You conversely seem to have a real zeal to promote HP on unrelated articles so I gues it's either plain old trolling, your a child-like geek who loves HP too much or perhaps you like lunch with the pretty sales lady a bit too much?

Any other "personal" incentives going on?

I don't really think the two above fit you however, I see so many people annoyed on these forums with your off-topic HP promotions it leads me even now to conclude your a HP sales related person, lying and presenting yourself as a teckie. Why the hell else would someone enthuse over a vendor with so much zeal? Maybe your an ex or current HP employed engineer faithfull to the cause?

I wonder how faithfull you'd be if hit with one of the layoff's?

As to your post : I also don't recall mentioning anything about billing systems, that should be a different class of iron from el-cheapo volume systems. Remind me where above I stated a T2 based system should replace true enterprise kit with expensive dynamic reconfig features for online component replacement as found in F6900's, 25k's, M-series gear and just as equally found in high end Power-series or HP superdome gear?

Oh yeah, I didn't, right.

The truth still stands that the Itanium gear was still floored in a recent price/performance trade off, the application of interest performed much better under simulated heavy load on the T2 gear. End of, no debate, I can't prove it to you any further here.

Sorry, your still a HP bigot and I'm still only commenting on observed real life system behaviour.

And I'm still an AC.

:-)

OK, we'll haggle, how's about meeting you part way ?

Both a HP bigot & a racist!

A bigot due to the rough dictionary definition : "One who is strongly partial to one's own group and is intolerant of those who differ" & a racist due to disliking people with a tinge of of purple or silver.

;-)

Hey, divot's a bit harsh, it was clearly meant in jest. A tiny step toward humour (ie: purple & silver) but fair enough about devaluing such a word.

Got the message, got me coat, time for me to F'offski....