Sun's latest Niagara and Rock details have reached El Reg, and they confirm that the hardware maker is up to some very ambitious stuff. First off, Sun looks set for the imminent release of its first Niagara II-based servers – the T5120 and T5220 systems. Customers will see 1U and 2U boxes, respectively, each with one of the " …
Too many threads....can software exploit it.....
I have talked to a lot of enterprises (mostly large banks, etc) and they think this is not as good as it might seem. Most enterprises tend to run their systems with common application patterns on one set of boxes, so application servers on one box, database servers on another, and so on.
There is some benefit in having a number of threads available, but for the common application pattern environments the software isn't that inherently multithreaded to be able to exploit that large a number of threads, and bottlenecks start to occur in caches, etc. So, a reasonable number of threads with a higher throughput of each is more akin to the software of the day...
I can't see myself telling Oracle with 11G or IBM with WAS to "make your software more multithreaded", as the overheads of synchronisation will become an issue.
Doesn't matter today. Just slice the box with many VMs. The more hardware threads and memory you have, the better. Each VM can run as a dedicated server. Think about a 256 thread box being sliced up into 64 virtual machines, each with 4 threads - and each VM will run efficiently because they would have dedicated hardware threads, no matter how slow each of those threads are.
Very imminent, you could try to buy one from Acclinet.
Not that exclusive a scoop is it ?
Sun Servers | SPARC Enterprise T5220
SPARC Enterprise T5220 System Options
Sun Fire T5220 System Options
Sun T5220 Spares & Upgrades Additional Upgrades & Options
Communication Sun Ethernet
Fibre Channel Sun Graphic
Removable Media Sun SCSI
T20-108A-16GA2C- SPARC Enterprise T5220, 8 Core, 1GHz, 16GB, 2x73GB, DVD
Multithreading push is on
Intel just open-sourced their TBB threading library, this will help to generate some activity in the thread-challenged development ranks. The common availability of multi-core chips is now pushing a lot more research (read: $$$) into low-level parallelism. While I don't see a breakthrough as imminent, I think we'll see more and better tools (like the Intel library) become more widespread, and as more developers get more experience with the technology, we'll start seeing some significant progress.
Re:Very imminent, you could try to buy one from Acclinet.
That's a T2000 part number dude, been around for over 12 months....... (check it out here: http://uk.sun.com/catalog/?n-state=http://catalog.sun.com/partpricing.xml?site%3dUNITEDKGB%26catalogue%3dFC%26segment%3dFC_R%26item%3dFC_SC_CAT%26group%3d2010%26fid%3d5150%26id%3d13308~~~G!07B24F0BFADE!WBPkexcnEpBE33lB~standard~ws-nocache~~@http://syndicator.sun-catalogue.com/ukunited/unitedkingdomcontext
Re:Very imminent, you could try to buy one from Acclinet.
That's a T2000 dude, been around for over 12 months.......
So no 2048 thread system actually exists, then?
So actually Sun is making a change in the code and the hardware doesn't exist? Yawn. Maybe a more apt title would have been "Next version of Solaris will be ready ". And 2048 teeny-weeny processing threads is as about as exciting as strapping 2048 toddlers together and telling them they are now a common giant intellect. Maybe SUN is designing a 2048-thread-capable router, but even with the addition of a maths unit the new chips will be pants in database performance even compared to old SPARC. As usual, lots of hot air SUNshine from Ashlee, and all about vapourware.
Banks in Terminal Decline or Joint AddVenturers in a New Wave of AIMastery?
"I have talked to a lot of enterprises (mostly large banks, etc) and they think this is not as good as it might seem."...... Given what they are protecting, it is no wonder that they are the most risk averse of enterprises, however given that their business IS just protecting their/your/our money, it is quite normal for them not to know of how good things can be whenever they do not work in the Processing of Information into Intelligent Designs which Create a NeuReal Artificial Intelligence .... and ITs CyberIntelAIgents Servering IT as a Global Mutual Intelligence for Present Chaos Replacement to Order. ..... for such has nothing at all to do with Banking and Everything to do with New World Order Systems.
However/Nevertheless/But it is surely Indicative that such a change as would be PROMISed in such a System is indeed required, for their reticence and doubt is unfounded and unwarranted in that any and all investments made in any new Process/idea/Business by the Banking System are never lost, and always just recycled within the System. Their doubts in protecting themselves/their clients wealth and not losing wealth are built upon a myth to support a dubious credibility which is halting Progress at every step of Innovations journey.
J'accuse ........ and suggest that they alter their Systems Programming to Support
with Finance rather than Hinder with Self Doubt, Sunnier Dispositions who Dare 42 Win Win.
Java app servers...
...are all highly threaded software, and are exactly what Sun had in mind for these boxes. Logically, isn't synchronization going to be less efficient if you are have a cluster of networked machines x each running y threads instead of a single high powered box running x*y threads?
I can think of at least one application for these monsters.
Server for online roleplaying environments. Or indeed for any multiuser online environment.
Can software exploit it?
No-one knows. However, what Sun *do* know is that if they compete on price in the "few-cores" market then they will get stuffed. Quite simply, Sun have no future unless this works, so betting the farm on "many cores" is not a risk.
Of course, as long as no-one follows, there's the prospect of Sun emerging in a few years time as the *only* computer company that has the OS and applications that can run effectively on what will by then be regarded just as a "modern" CPU.
I didn't say you could buy one...
... only that you could try. ;-)
Look a bit more closely and you will see how the site doesn't quite hang together. The links for system options point you to the T2000 entries but it does helpfully confirm the model numbers and the like.
Sun Servers | Sun SPARC T5240
Sun Fire T5240 System Options
Sun T5240 Spares & Upgrades
Additional Upgrades & Options
Sun’s new cutting edge servers utilizing Niagara chipsets are highly anticipated and are sure to live up to the hype. The SPARC Enterprise T5240 is a compact 2U server featuring a one-way Niagara 2 chip. As with all Niagara servers the SPARC Enterprise T5240 will have extremely low latency due to a direct integration of key components onto the chips themselves. The Niagara chip in the SPARC Enterprise T5240 excels at processing multithreaded applications and can handle hundreds of threads simultaneously. Exact numbers for the Sun T5240’s RAM and disk configurations are still sparse at this time, but it should be just as impressive as the new SPARC chip. For more information on the Sun Microsystems SPARC Enterprise T5240 contact Acclinet today.
Logical Domains is not ready, IMO, to be trusted with multiple production workloads. There is still the problem with the control domain being a single point of failure for the whole system. Another issue is the way that the threads are allocated to virtual machines (dedicated to specific VMs) forces you to plan for peak workloads on each VM. This means you're purposely leaving CPU processing power unused in times of typical workloads rather than allowing the hardware to give it to starved VMs (e.g. defeating the purpose of virtualization).
Re: Use Virtualization
Sun has other "virtualization" technologies you know? They are called zones and containers. The latter is a zone with resource management. If you don't know about the technology then i suggest you read up on it. It does and will use all the threads/cores between the zones without resource management by default. Since each zone is still using the same kernel and there are no restrictions unless applied by the sysadmin every ounce of resources will be used when needed. To get the most out of this you would, of course use FSS and highly threaded apps
As for production use, they work very well and since i only have to patch 1 box in a sparse root model my time is now spent reading El Reg instead of M$/VMware patch/reboot VM hell. As for single points of failure, wouldn't you run you apps in some type of load balanced cluster? JBOSS and apache/tomcat comes to mind as good examples.
Comment on the threaded monster,
It may not work for most but i do see something like BI taking advantage of it. I could see large organizations reaping the benefits sifting through terabytes of data with hundreds of users doing analysis all at the same time.