Sun takes four-socket Victoria Falls Sparc plunge Sun Microsystems and server partner Fujitsu have announced a four-socket mid-range server based on the Victoria Falls Sparc T2+ processor, following weeks of strong hints. Fujitsu did most of the engineering on the Sparc Enterprise M servers sold by both companies and based on the quad-core Sparc64 VII processor. However, the …
Sun, still innovating god bless them. No matter what you have to admire the pure enginnering behind lot of what they do. Will it be enough only time will tell but I do still believe they are one of the few companies in tech with new ideas.
Now cue Pavlov's dog ... Matt Bryant.
"As T5440 is a NUMA architecture, memory latency should be considered when running database applications that are very sensitive to memery latency" from Sun's own Blog site
Oracle DB is $1.1M and Oracle WebLogic is $600K on the system plus $374K per year in Oracle maintenance.
It's also bad the chips don't support heavy threads that a database needs and Sun has no plans to go past four chips.
The fact that Sun removed two memory controllers per chip for the T2+ processors also hurts the scalability.
Interesting that Fujitsu is manufacturing the systems. Seems Mini-Me is progressing with the exist the hardware business.
Whisky Bar
Just checked out spec.org. The SPECjAppServer2004 results shows the four socket T2+ has less performance than three T2+ processors.
One chip does 2000 JOPs.
Four chips do 5836 JOPs.
In today's expensive software market one would expect at least 90% scalability in a rigged benchmark. Customers should beware of the performance and resulting TCO.
Who is Anonymous Coward? Man you're so down on Sun! What did they ever do to you? Sure the T5440 isn't a panacea, but it's really nice kit and finally a good migration for legacy Sparc and Solaris. Give Sun credit for some really awesome engineering - can you get a 256 thread (in h/w) box anywhere else? Whatever you think, sales numbers are saying that people like buying T2 and T2+ boxes. Finally some Sun R&D starts to pay off. Let's hope Rock doesn't flop.
The article states that:
"The Sparc T2+ chips have a floating point math unit for each core, plus an on-chip cryptographic processor and 4MB of L2 cache that is shared by the cores."
This is essentially an understatement... Each core has it's own dedicated cryptographic processor which means 8 crypto-processors per T2+ and 16 for the whole server described.
The 4 Megabytes of cache "shared by the cores" is also something of an understatement because there is a cross-bar between the L1 cache of each core and the 8 separate L2 cache banks. The cross-bar is far more quick and efficient allowing threads to jump from core to core without having cache flushes. Finally, each core's L2 cache talks directly to it's own memory controller which in turn talks to it's own dedicated DIMM slots. In effect you have 8 ultrasparc computers each with a single core that has 8 simultaneous threads and it's _own_ crypto processor, L1/L2 caches & DIMM memory.
This server throws down 32 of the above into a 4U box and up to 512GB of memory.
HELL YES.
http://www.sun.com/processors/UltraSPARC-T2/datasheet.pdf
Daniel Chapiesky
The comment AnonymousCoward quoted was totally out of context and does not tell the whole story... see:
http://planets.sun.com/ISVe/group/ISVE/
"As T5440 is a NUMA architecture, memory latency should be considered when running database applications that are very sensitive to memery latency. Fortunately, the threaded design of UltraSPARC T2 Plus chip acts to minimize the impact of memory latency. DB2 also supports the optimal placement of a database partition's processes and memory, which means a database partition's processes and memory can be restricted to an Locality Group (lgroup), so there are lower memory latencies and increased performance."
Your not going run oracle on one these boxes... They have no local storage beyong 4 disks and you can't even get a good raid array with that many. They do have 8 pci slots and 2 10gigE ports along with 128GB for about $140K.... Load the thing up with network cards and it can be the web server/light thread server from hell....
factor in the power and cooling needed for 14 dual xeons to compete with this box and the TCO drops enormously.
The whiners about oracle licensing should get 1400 ipods and shut up.... The TCO of oracle is separate from this sun server and staying on an older box because of oracle licensing means your IT department can't support your applications and has no idea about what TCO means... It includes disaster recovery, support salaries, and little things like points of failure which increase exponentially with each linux server you add.
Give me four of these sun servers and I can replace 4 RACKS of equipment and fire the 2 slack employees that I would otherwise need to install the latest ubuntu pack, which BTW breaks our applications everytime there is a version upgrade.
IN ADDITION.... get your sorry asses off apache and go to haproxy/nginx/& yaws.... NICE THIN THREADS THERE and the T2 with partitioning will stomp anything you throw at it.
God I hate people who think they know what TCO is...
"What Sun has said is that is relation to its own Sparc T platforms is that is could deliver about 1.9 times the performance when doubling up on the processor count in the Niagara machines from one to two sockets, and that moving from two to four chips it could deliver about 1.8 times the performance."
I think I'm almost sure I know what was in the author's mind.
Have you not heard of virtualization and consolidation, the T2's are what they are and not half bad, I'm running five of them in an Oracle environment, and yes if you just slap your software on to the machine and expect it to run efficiently and cost effectively you will be disappointedly, and Oracle will quite happily charge you a s%6 load of money for your trouble.
However if you use Logical Domains, (LPARs in IBM parlance), to allocate the amount of CPU you require to processor power you need, and you tune your application to make best use of the number of threads and cores you have, then you won't be disappointed, Larry will, because he won't get lots of money for nothing. Oh, and make sure you get the I/O right.
If I have a criticism of the T series boxes, it's that, by default you start off being I/O bound, but the history of computing has always been about shifting between architectures that have cured one problem only to be slowed by the next.
These systems are perfect for consolidation, LDOMS and/or Containers; a big pool of threads and memory in a 4U enclosure .
Listen just because Oracle are still clinging to a late 20th century licensing model, that's not going to stop hardware evolving. Eventually as the core count increases on AMD and Intel also, Oracle will be dragged kicking and screaming into the real world.
Well if you go with "Named users" you will have a minimum of 800 users. 25users/core * 32 = 800users. Still over $1M with 3 years maint.
As the prior person stated the SPEC jbb benchmark has over $2M in Oracle software.
I wish Sun would stop pushing these boxes as Oracle solutions. It is deceiving to customers. The max of four harddrives and outrages software pricing make these only good for free software stacks.
And who was the idiot that did not use MySQL for the java and SAP benchmark? Hello?
Another $1B wasted?
"Let's hope Rock doesn't flop." Found this in googling Sun Batoka
www.geocities.com/sunsrockchip
Oracle are always slow on the uptake with their licencing models, they basically want to charge what people will pay, when dual core first came out they didn't change the models and sun was very happy, our E25k looked cheap then they started the 1.4x charging model and it wasn't so cheap, they'll change it again, and again. No doubt M$ will also have a rethink, current charging is per socket, but as more cores appear they'll change it (maybe Windows 7 will have a new and exciting pricing model?)
Oracle licensing for Sun gear seems to have got prohibitive since the MySQL acquisition.
They charge 50% more for a SPARC core than an x86 core, and treat T2+ cores the same as SPARC64VI+ cores. With all the press about wins against each other, I can't see Oracle wanting to make Sun platforms attractive any time soon, they want shops to reuse their licenses from their old SPARCs on new Oracle Linux boxes from whoever....
Pavlov's dog reacted to the sound of a bell. No bells here unless you mean the Clanging Chimes of Doom (TM) that resonate with Sun's stock price - maybe it's time for SunAid? "Buy a box that can't do the job, because it's charity...."
Sun needs these boxes to keep old Slowaris customers buying Sun kit, for those old Slowaris 8 and 9 apps. The problem is majorityof those old servers were running database apps like Oracle. You may recall that Sun used to run around bragging about how it was Oracle's number one partner? Well, that was mainly on old SPARC with Slowaris 8 and 9, and a lot of those customers have already given up on Sun and moved to other commercial UNIX flavours or the dreaded Linux. Sun is desperate to retain that business as they know they have no chance of fending off x86 in the edge server arena, no matter how Sunshiners like Daniel Chapiesky sprout on about TCO. The fact is webserving is a Wintel/Lintel domain because you don't need a horde of expensive Sunshiners to run a cheap webfarm of Wintel or Lintel, you don't get stuck with a Sun-only solution with limited options, and you don't get stuck with Sun's rapidly deteriorating support.
And as for not testing enterprise Slowaris 8 and 9 apps but just dropping them into containers on Slowairs 10 - are you nuts!?! Many of these apps are still running because they are business critical or simply too difficult to port. There is no way a serious enterprise bizz would try droppign them into containers without a thorough round of testing, which means more cost which seems to never appear in those nice Sunshiner TCO studies. And that's long before we start looking at the real performance hit all Niagara systems suffer from being unable to handle heavy, single-threaded apps (like Oracle!), which means you won't consolidate many onto any Niagara box at all.
Oh dear, maybe it would be a better idea to let sleeping dogs lie in future....
I’m a Linux geek, have been since 1993 (Slackware!). All of SmugMug’s datacenters (and our EC2 images) are built on Linux. But the current state of filesystems on Linux is awful, and it’s been awful for at least 8 years. As a result, we’ve put our first OpenSolaris box into production at SmugMug and I’ve been pleasantly surprised with the performance (the userland portions of the OS, though, leave a lot to be desired). Why OpenSolaris?
http://blogs.smugmug.com/don/2008/10/10/success-with-opensolaris-zfs-mysql-in-production/
"Sun's stock price"
Was rising about 20% yesterday...
"TCO"
Funny, last time I worked for IBM, we charged the same price for running Linux as for running Solaris (and HP/UX and AIX)
"you don't need a horde of expensive Sunshiners to run a cheap webfarm of Wintel or Lintel"
Last time I checked, we payd the same salary to Linux guys as to Solaris guys (and HP/UX and AIX as well). But it was very hard to find any HP/UX admins. It seems like they have left the sinking ship called Itanic. Maybe we have to offer the candidats a higher salary ... err...
"not testing enterprise Slowaris 8 and 9 apps but just dropping them into containers on Slowairs 10 - are you nuts! Many of these apps are still running because they are business critical or simply too difficult to port."
Hey, have you heard about this "new" Solaris feature called binary compatibility guarantee? I can tell you, it requires not that much testing to move these applications to Solaris 8/9 containers. Maybe HP should start their copiers to implement something similar for all those legacy PA-RISC applications around, that are business critical and simpy too difficult to port.
"Sunshiner TCO studies"
Let's see what the TCO is for porting PA-RISC applications to Itanic... maybe not...
"single-threaded apps (like Oracle!)"
Oh, I thought Oracle was good in running things in parallel, ...
Pavlov, play dead! ...Good boy!
Ah Matt the old "Slowaris" line again. We're just moving a web pool off RHEL4 onto Solaris 10, we're getting about 10-12% better performance - same hardware. You know what, we've also run the Linux Apache binaries in a Solaris 10 BrandZ Container and they were still around 3% faster than native RedHat on the same hardware.
And when it comes to churning this hardware I'd say we'll be taking a look at the T2 systems too.
"..."Sun's stock price"
Was rising about 20% yesterday..." A lot of shares rose initially on the news of the Bush recovery plan, but a lot slid back later in the day, including Sun, and the NASDQA closed down. Meanwhile, Sun's stock is still under threat of being delisted. Nothing like the whole story to ruin a Sunshiner's party!
"....Funny, last time I worked for IBM, we charged the same price for running Linux as for running Solaris (and HP/UX and AIX)...." I suspect you wre one of the overpaid oxygen wasters IBM let go then? In the real corprorate world, where there has alwyas been the split of the lower-paid Wintel team and the higher-paid UNIX team, I have seen management keen to push new products onto Wintel as they know it means they can reduce the number of expensive UNIX admins. The smart UNIX admins have seen this coming for years and have been learning Linux. And as for IBM hosting prices - don't make me laugh! I have seen massive differences quoted for different platforms. And when we did try them we had the disconcerting case of Global Services calling us, the customer, because they didn't have an hp-ux resource available to make some changes but had taken on support for a 24x7 hp-ux solution! I can't think of anyone less qualified than an ex-Global Screwups employee to comment.
"...It seems like they have left the sinking ship called Itanic...." That would be a dig at the Hp Integirty servers, then? The server range that is the only enterprise server range that is growing year on year? And don't bleat on about Niagara, it's not enterprise, it's barely edge server, and not a real replacement for UltraSPANKed or any other RISC server, which is why Sun has to rebadge FSC SPARC64 servers and pray Rock makes it out alive. You may no have noticed (not hard to understand why with that Sunshiner Blinfold (TM)), but Integrity servers can run hp-ux, OpenVMS, Windows, Linux, or even Solaris on Transitive, and can mix all the above in electrically isolated partitions (that's real highly-available virtualisation, which Sun can't match). And I think you'll find the majority of the hp-ux contractors are happilly working in the City where hp-ux is still the number one UNIX choice. It's not surprising you can find plenty of Sun admins out there as there are plenty of them out of work.
"...Hey, have you heard about this "new" Solaris feature called binary compatibility guarantee?..." Yes, I have heard of it. When we asked Sun to provide penalties into a support contract for the "binary compatibility gurantee" they wouldn't. So if Sun don't even have the confidence to back up their claims with their money why w=should I bet the company's money on it?
"....Let's see what the TCO is for porting PA-RISC applications to Itanic... maybe not..." Well that's because hp-ux 11i v3 has full binary compatibility across Itanium and PA-RISC, and I can tell you for a fact that if an 11i app hasn't been hard-coded with specific library links then it will usually port without recompilation. Which measn the TCO costs for migration are usually very low. Most such porting tasks consist of running the code through hp's transition tools, which require very little in cost, and provide a reliable indication of where any problems lie. Compare this to Sun's approach of cross your fingers and hope and you may see why admins in the real world will tell you hp-ux migrations of business critical apps are a piece of cake compared to Slowaris or AIX.
"....Oh, I thought Oracle was good in running things in parallel...." Yes, it is very good at doing full-sized heavt threads in parallel on proper CPUs like Itanium, Power, Xeon, Opteron or even SPARC64, not on wheiner latency engines like Niagara, hence why Oracle performance is awful on any Niagara system, being outperformed by the avergae Wintel whitebox for a fraction of the price.
Which is why, if you want to see dead, just watch Sun for a bit longer, you won't have too long to wait.
RE: Ding, Ding - check the market figures, you will see that Lintel is eating up the Sun marketshare for a reason - they don't believe Sunshiners like you.
Lots of these comments read exactly from other vendors scripted marketing hit pieces. Too much passion about being anti-Sun to be believed. And all with no facts. Maybe your job is at stake if you don't bash Sun? Really, people see through this bad marketing.
While I posted anonymously, I will will very clearly state that I work for Sun.
Why not look at some facts on features, great scaling, performance a very good set of information from a wide variety of Sun bloggers (who are predominantly working engineers). Start with Alan Packer's (Sun) blog.
http://blogs.sun.com/allanp/entry/sun_s_4_chip_cmt
Seems a dirty little secret in the computer industry is not measuring power on all benchmarks.
I think Sun should be congratulated on publishing measured watts on the T5440 benchmarks.
These benchmarks are full configurations run at full utilization with none of the configuration games played by other vendors on "power" benchmarks. Will the Reg ever dig up the dirt on these configuration games and expose that silliness?
I work for Sun.
I've got sun's, ibm's, hp's & no name's. If I had the budget I'd drop everything except for the sun's. Why...? Consistency. Solaris/OpenSolaris on the smallest to the largest. Stable releases and containers for testing. Oh... and someone to sue....
Yeah I said it. Sue.
When it comes down to it the lawyers in your company drive many decisions which your CIO is told to implement. I love linux and it has its place. I hate windows and it still has its place too. Why do companies with millions of dollars of revenue use windows and solaris... We can sue the hell out of the companies that make them.
Does that mean we will? Does that mean we can, in actuallity? No. But we can say to our stock holders that liability and exposure are minimized by going with these vendors.
The T5440 is clearly a compute server. It also appears to be a swiss army knife waiting to be outfitted with the tools you want since it has 8 pci slots... Do you want a visualization server? Throw 256 gigs of ram in it and 4 video cards.... now you can drive monstrous displays with ONE BOX.
Do you want a storage virtualization server.... throw 256 gigs of ram in it, 6 dual channel FC cards, 2 8 port 1gigE cards or 2 10gigE faceplates.... and you got a kick ass storage virtualization server.
Do you want a web server from hell? Do the above with forward facing and rear facing gigabit cards.
Oracle pricing? screw oracle. Stored procedures is the only reason anyone stays with oracle. Rewriting them for mysql is a bitch and no one wanted to do that because until recently mysql was impossible to sue if someone screwed up.
Sun guys? Listening? You got mysql now. Your customer's lawyers will now give a go ahead on using mysql because of it. You guys write an Oracle to mySQL stored procedure converter... and you can throw down cost savings across the board for your customers.
For God's sake.. don't let your server sales people kill that idea cause it would effect oracle server sales which they may get a cut of... No in-fighting.
Anyway.... The T5440 is a bad-ass box and I think it really is just a prelude to rock and T3.... for working out the chassis/cooling/power etc...
I am not a Sun employee but I's got's a bunch of 'em.
I am a linux kernel hacker and have C++ in my kernel... do you?
I am a windows programmer... and I still use visual studio 7 cause .NET sucks.
And finally... I pay the bills and own everything. Business is first and always first.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
