SPARC64 Strikes Back on Super-Computing

It would appear that Fujitsu has learned a lot from Sun's CMT processor designs and is testing out its new SPARC64 VIIIfx or Venus processor which is an 8 core SPARC64 chip that can perform 128 Giga-flops! Here are some interesting points about this processor:

• 8 core SPARCv9 with extended SIMD extensions for supercomputing, known as HPC-ACE.
• Dedicated L1 data and instruction cache for each core.
• Shared 6MB L2 cache
  
• 45nm CMOS process on 2 cm squared die
  
• 2x the number of transistors as the SPARC64 VII, yet only one third the power consumption.
  
• Integrated DDR3 Memory Controller
• Hardware optimization for non-parallelized traditional workloads
• 128 Giga-flops performance per chip, more than 2.5 times the fastest Intel processor.
• Liquid cooled!

Fujitsu is first targetting this processor for super-computer applications into a 3D torus arrangement that will allow for over 100,000 nodes to be connected together and deliver around 12.8 Peta-flops! This will out perform the FX1 SPARC64 VII super-computers which run at 40 Giga-flops per chip. The FX1 is only sold in Japan, you can read more about it at this link.

Here is a picture of a Venus node:


As you can see, the front of the module is on the right-hand side where the latches are. You can also see what look like the liquid cooling connector on that side. This would suggest that the node plugs into a larger chassis and that the I/O components are on a separate module. I would not be surprised to see that the frame uses a custom switch-less fabric on the back-end, with mainframe level RAS, and SPARC Enterprise M-Series features.

It remains to be seen when this processor will make its way onto the M-series or a future SPARC product line for Fujitsu and Oracle/Sun. However, with the investment from Japan in deploying this in super-computers and Fujitsu's focus on making traditional workloads benefit from this new processor, it is very likely we'll see Venus based SPARC servers in the near future. The M-series was designed to handle several generational upgrades. It has already gone from Dual-Core to Quad-Core. As such, an 8-core upgrade is only logical. How this will play into Sun's ROCK processor is in question as well. It would appear that Fujitsu has indeed learned a lot from Sun around building multi-core and multi-threaded processors. ROCK and next-gen Niagara processors may be more focused on general purpose computing, while Venus is focused on large number crunching super-computer tasks.

Mergers and Legal Dribble

The SEC filings on the merger are available for the public to review. The document has lots of interesting details about the merger process, conditions, etc. Reading through the document provides an interesting lesson in how mergers are designed and how things get sorted out. Items such as severance pay for the executives, beneficial stock holders, waiving of rights, etc. are all very fascinating to read through. It provides an interesting glimpse into the legal process and how companies organize such activities. I was not surprised to see my old employer BGI holding a significant amount of securities in the company. I think those who hold on to their stock and enjoy the gains from the conversion to Oracle common stock will be happy over time.

Of course some of the other interesting bits of information surround the history behind the merger itself. Some of the other parties who were interested, have been masked over. Reading through the document, it's obvious the other suiters were HP, Fujitsu, and of course IBM. I never thought that Sun would be a good fit for HP, other than to kill off the competition, considering HP's rather bloody and unfruitful past mergers. Fujitsu has done worse in the server market space outside of the Asia region, and has such a limited market in the US, it didn't make sense unless they were willing to commit to the US market. While the R&D mentality of Fujitsu would have brought something to the table with Sun, they seem to suffer the same sales and marketing issues Sun has been unable to address over the years. Ultimately, I still think the Oracle merger makes the most sense seeing how Oracle has strong ambitions, sales, and marketing power. The vision of supplying the whole solution stack makes increasingly more sense as one looks at the IT landscape. Oracle stands to change the game and go headon with IBM and HP.

Speaking of a changing landscape, Rackable is merging with SGI and taking on its name. SGI has been through some rough times over the years. I remember when I first lived in Silicon Valley seeing the huge campus for SGI and thinking they must be doing well. It's amazing how fast things can change.

And speaking of Oracle, it is buying out Virtual Iron which will give Oracle access to some interesting Xen management tools. It might also be interesting to see how Oracle will cherry-pick components from Oracle VM, Virtual Iron, and Sun's xVM. Oracle is definitely tooling up on components for its stack. Considering that Oracle will own Dynamic Systems Domains, Logical Domains, Solaris Containers, xVM Server, VirtualBox, Sun Rays, etc. through the Sun merger, they will have the widest virtualization portfolio in the industry. A solution for each situation and platform. Something to think about:)

More Rumblings and Solar Flares

There has been some talk lately over Sun's SEC filings which list some interesting turn of events:

• Sun shareholders acting to block the Oracle take-over.
• Violations of bribery laws

The first one, I am not surprised at since Sun's management has indeed failed to meet their responsibilities to the shareholders for manning the ship and keeping the profits up. I'm not sure how this will help, since a failure to complete the acquisition would probably place Sun into further uncertainty and force it to go private or bust. Hopefully this will get resolved peacefully. But I do understand the frustration and anger shareholders must have for being let down constantly over the past 8 years.

The second item though, is indeed troubling and could hurt Sun's position in the GSA program. Idealistically, business should not resort to such tactics. But that is not the way of the world. Recently news of the EU filing anti-competitive charges against Intel for paying manufacturers to not sell AMD processors demonstrates this clearly. Not to mention that Microsoft is having its own issues abroad as well. I've seen enough business deals change direction due to vendors using such tactics or discounting hardware to bargain-basement prices just to make the sale, but charge you an arm and a leg on support and licensing. The sad thing is that customers forget that they have choices and need to leverage that to not only get a good price, but to have a good business partnership based on trust. Hopefully those responsible for the bribery issues will be dealt with. Atleast Sun has been honest about this issue and is helping the authorities.

Ecoviv Beta Testing Wand for LDoms

The company Ecoviv is beginning to beta test their management product for LDoms, called Wand. This product was previously known as Virtvue, but has some significant enhancements:

• Web based remote management of LDoms
• A robust three tier architecture that leverages Solaris’ Cacao Agent
• Support for LDoms 1.1
• LDoms Life-Cycle Management
• Live Migration
• Provisioning of the Logical Domains Manager and of Guest Domains
• LDom Search and Discovery
• SNMP Support

This is great news and I'll hopefully get into the beta program. This would be a fun product to write an article or white paper on.

Oracle Confirms Commitment to SPARC

Reuters did an interview with Larry Ellison over the acquisition of Sun that has some great news. In the interview the subject of software vs hardware came up. He confirmed that Oracle will retain Sun's hardware business and increase development of the SPARC processors as important to Oracle's strategy for delivering a complete solution stack. This is great news for SPARC customers to hear and understand. Larry Ellison even specifically stated that Oracle wants to compete directly with IBM in the data center for dominance. Oracle is aware of the value of the SPARC64 and the UltraSPARC CMT lines as they are power efficient, scalable, and heading in the right direction. I highly recommend this article for customers, users, engineers, etc. to get the message out that Oracle is committed to Solaris and SPARC. I think with Oracle's focus on competing with IBM Power systems, we'll see some significant focus on ROCK and the next generation CMT systems. Hopefully, we'll see SPARC kick Power7 in price and performance.

Solaris 10 05/09 Released

The latest update to Solaris 10 has been released. There are new features and fixes for Solaris Containers, SMF, IPSec, 10GbE, iSCSI, Infiniband, SunSSH, and even SunVTS. No major updates on platform support or virtualization. Seems this should be a fairly stable update.

In other news, it's encouraging to finally see builds of the Indiana or OpenSolaris.com flavour for SPARC. Unfortunately, this release can only be installed by the new AI installer over the network and can not be booted from CD/DVD or through Jumpstart. Another major limitation is that Xsun is not included, which limits framebuffer support significantly. Until recently, we only had Solaris Express or one of the other distributions such as MilaX or MarTux. You can check these different distros out on Genunix. Overall, it is encouraging to see the number of distros for OpenSolaris grow and mature. Hopefully, the Indiana codebase and support will mature quickly on SPARC. (We can only hope that Oracle will see the value in making cheap UltraSPARC-T2 workstations).

Switching from SPARC to and Opteron

A couple of weeks back, I bought an Ultra20 off of Ebay to replace my aging Sun Blade 2000. I've been pretty happy with my Sun Blade 2000 over the past few years. However, it was beginning to worry me that due to the closed source drivers for the Expert3D card, I would eventually be left in the dark with OpenSolaris and probably Solaris 11 when it comes out. And due to the fact that Sun no longer makes a SPARC workstation, it seemed like it was time to move on. I have for years run Solaris Express and recently OpenSolaris on VMware and VirtualBox on my work laptops. I do miss having OpenBoot and the Sun Logo lit up on my workstation, but I'm very happy with the transition. Everything obviously runs faster and getting use to IPS has been pretty easy. I did do a few upgrades to my Ultra20:

• Opteron 180 2.4Ghz Dual-Core
• NVidia GeForce 8400 GS PCI-E
• Sony DVD+-RW Drive
• IO Gear USB->Serial Adapter (Still need serial access to gear I test)

While it's not the fastest x64 system out there, it was amazingly cheap on EBay and works great! Here is a picture of the system before the upgrades:


Here is my old system, which I'm selling on EBay:


Here is my new desktop on the Ultra20 using the Avant Window Navigator which gives a kinda MacOS X feel and is very snappy:


I've taken the time to recompile many of the apps and games I had on my Sun Blade 2000. I'll have to see if I can package them up into IPS and distribute them for other folks. Overall, the transition has been fairly painless and a lot of fun. At somepoint, I'll have to upgrade the motherboard, CPU, and RAM to a newer Opteron generation or a Phenom II x4 so I can do xVM Server testing. Now all I need is my own T1000 or T5120 for doing LDoms on.

Fujitsu Out-Sourcing to TSMC

Well according to reports, Fujitsu is divesting itself of semi-conductor fabrication and out-sourcing it to TSMC. This is not as surprising if you consider the amount of consolidation in the technology sector. This is something that has already happened in hard drive manufacturing where IBM sold out to Hitachi and in July Toshiba will own Fujitsu's hard drive business.

What I find really interesting is that Sun had already selected TSMC to fabricate the ROCK or UltraSPARC-RK processors and the next generation Niagara processors. Fujitsu was already producing the SPARC64 processors for the SPARC Enterprise M-Series servers. It'll be interesting to see if TSMC will be manufacturing the SPARC64's as well in the near future. It's unfortunate that Texas Instruments will not build a next generation fab for the Niagara chips, but it is an expensive investment. Even AMD has spun off its chip manufacturing business due to the expenses involved. This leaves Intel, Samsung, TSMC, IBM, and the Foundary Co. as the major chip manufactures for processors.

While there was speculation that Fujitsu would buy Sun, considering how Fujitsu has been divesting itself of manufacturing hardware, it didn't seem to make a lot of sense. The SPARC Enterprise servers are not manufactured by Sun or Fujitsu, only certain parts are. The rest is manufactured and assembled by Flextronics. Even HP sold off its share of the Itanium and Alpha business to Intel. To make things even more interesting, with the rise of SSD storage in servers and storage arrays, the future of the storage market will have little to do with spinning disks anymore and a whole lot more to do with chip manufacturing. This is something that Intel and Samsung are investing heavily into and even Sun has worked closely with both to further this trend.

The Niagara processors have been ahead of the general purpose CPUs in the multi-core and multi-threaded game. It is interesting that since the introduction of the T1000/T2000 servers, Intel and AMD have stepped up on the creation of dual, quad, etc core processors. Even IBM has realized that Sun was right and is now having to eat its own words against CMT as it develops the Power 7 processor featuring multi-cores and threads. Another trend in this area is the inclusion of features such as memory controllers and large caches into processors. Of course the UltraSPARC-T2 and UltraSPARC-T2+ processors have additional features such as cryptographic units per core, on-board PCI-E controller, on the UltraSPARC-T2 a 10GbE controller, and on the UltraSPARC-T2+ SMP interconnects. The concept of a "system on a chip" will continue to expand and become prevalent.