Sunday Oct 05, 2008
www.flickr.com
|
Malhar Anaokar
SeChang Oh
Ken Pepple
Divyesh Shah
Takashi Shitamichi
Saday Tiwari
Today's Page Hits: 70
Sunday Oct 05, 2008
High wire acrobats frequently walk the wire without a net, but chips seem set to trump them by communicating without wires. Proximity Communication continues to feature regularly in the news as it matures. The Sun Web feature Contrarian Minds describes Jack Cunningham's work this month. Cunningham leads the team working on an advanced packaging solution for the technology, focusing on placement accuracy and power supplies.
Alignment of chips is vital to Proximity Communication - if the transmitters on one chip do not align perfectly with the receivers on the other, the coupling might not be strong enough for communication to take place. Jack and Ashok Krishnamoorthy have come up with a technique involving pits on one chip and what they call microspheres on the other. Bring them to within 200 microns of each other, let the chip with the pits go, and voilà, it rolls into perfect alignment with the other chip. A pit and ball approach, if you will.
The article also describes Cunningham's work in the design of how the chips receive all the power they require.
Some of Krishnamoorthy's other research is described in a June 2008 Contrarian Minds article - using light to communicate between chips turns out to be a pretty smart idea. Such research resulted in a DARPA contract for the Ultraperformance Nanophotonic Intrachip Communication project, and The Register reported recently on the work that Sun will be doing along with Kotura Inc for the project.
A September 2005 Computerworld article went over some of the history of Proximity Communication research at Sun.
Tuesday Jun 03, 2008
InfoWorld honours extra-ordinary technology leadership by releasing their list of Top 25 CTOs of the year. This year's list includes Greg Papadopolous, CTO of Sun Microsystems Inc. Congratulations!
The article accompanying the citation mentions a slew of innovations emerging out of Sun's Engineering - Project Blackbox, Systems based on Chip Multi-Threading, memory and networking subsystem advances and OpenSolaris.
Reliance Communications uses the Data Warehouse Appliance
Greenplum and Sun Microsystems Inc announced yesterday that Reliance Communications Limited has deployed the Datawarehouse Appliance powered by Sun and Greenplum to support applications ranging from legal and regulatory compliance to call detail record analysis.
The press release went on to describe the impact of the deployment :
.. The time required to turn around an ad hoc request for detailed call records from the historical call detail record (CDR) database shrank by over 80 percent in most cases, from five hours to less than an hour, and ... the average time to load a day’s worth of data [reduced] by over 80 percent, from two hours to under twenty minutes.
The Datawarehouse Appliance combines the parallelized shared-nothing Greenplum Database with the innovative Sun Fire X4500 Data Server, Solaris 10 and ZFS technologies to deliver scalable data capacity and query throughput, and attractive price/performance.
The Appliance configurations include a Master/Query Dispatch Server hosting the Greenplum Database master segment instance to provide user connection services and query optimization/dispatch. The data segments are distributed across a bank of Data/Execution Servers. The tightly integrated storage and compute elements in the SF X4500s are particularly well suited to datawarehousing environments.
Tags : BIDW Datawarehouse Greenplum India
Thursday May 29, 2008
The Austin American-Statesman features the Ranger in an article describing how Austin has begun claiming a bigger piece of the supercomputing action. The Ranger is the most powerful general purpose supercomputer in the world, funded by the U.S. Government's National Science Foundation, and developed by Sun in conjunction with the Texas Advanced Computer Center.
TACC, Sun and AMD collaborated with the Austin Institute for Computational Engineering and Science, Arizona State University and Cornell University in the application development, deployment, management and training strategies around the infrastructure.
At the heart of Ranger is the Sun Constellation System - a combination of ground-breaking compute, storage and networking components from Sun. The American-Statesman article is accompanied by an interactive graphic illustrating how you get from the AMD Barcelona processor to a 3936 node cluster. Peaking at 504 TeraFLOPS, Ranger is a huge step in the drive toward Petascale Computing.
Tuesday May 27, 2008
Sun Microsystems is one of 74 brands (chosen from 987 entries across 129 categories) declared Business Superbrands in 2008 by Superbrands India. This achievement underscores the reputation we are honoured to have across the Enterprise and Developer communities in the country.
Way to go, Sun India!
Tags : Superbrands Sun India
Thursday May 22, 2008
Steve Staso sent out an e-mail that alerted me to the latest addition to the Sun.COM domain - Channel Sun : infrastructure that allows Sun employees to post videos on-line. The first videos are already up - quite a few interesting interviews conducted by Sumaya Kazi at JavaOne, and a useful Video Tips and Tricks show by Steve Ewertz.
Channel Sun joins blogs, wikis, podcasts, and the various communities that Sun employees participate in and contribute to. I also bump into colleagues (and the content they generate) on Facebook, Twitter, YouTube, FriendFeed ... Did I say colleagues? I might never have met quite a few of them, but they are friends now.
As for Channel Sun, Lights, Camera, Action!
Tags : JavaOne Sun UGC Technology
Saturday Feb 16, 2008
I am not much of a basketball fan, so apologies for a misleading post title (attributable to the partial attention I paid to ESPN India's telecast of a Maryland-Duke game). I managed to install the Indiana Developer Preview 2 OpenSolaris distribution on my TigerMacOS X 10.4 laptop using VirtualBox yesterday.
A 20MB distribution that allows me to run Indiana DP2 on top of MacOS X sounded too good to be true. Thanks to simple VirtualBox configuration and Alan Burlison's advice on guest OS networking, obtaining a connected Indiana VM on my Macbook Pro was a breeze. Entertainment by the redoubtable Christopher Hitchens during configuration possibly helped too.
I used the PCN driver from the Solaris Express Community Edition for networking as Alan suggested, but his post also contains instructions around (and an ISO image of) Masayuki-san's free NIC drivers for Solaris. I set 1GB of base memory and 32MB of virtual memory for my virtual machine.
The idea of a layer managing and provisioning resources for applications and abstracting those resources from the environments in which the applications run is not new, of course. However, virtualization seems to have come into its own lately. The innotek acquisition is the most recent milestone in the history of our virtualization portfolio - a history that has looked very busy in the last decade or so with my favourite highlights being Dynamic System Domains (a capability that has evolved several generations since the E10K days), Solaris Containers, Sun Ray thin clients, Chip Multi-threading, Secure Global Desktop, Logical Domains, ZFS, the Storagetek acquisition, Project Blackbox and Sun xVM.
There is no particular need for the screenshot that occupies much of the real estate here, but it saves you a thousand words.
Tags : virtualbox macos opensolaris
Wednesday Oct 10, 2007
Innovations in Sun's chips, interconnects, operating system and data management products are emphasizing the fact that Sun is a Systems company, and not just a hardware company, or a software or a storage one. At the Customer Engineering Conference this year, the buzz is around products that combine these components into a system more valuable than the sum of the parts.
Sun Microsystems spends approximately two billion dollars annually on Research and Development. Over the past couple of days at CEC, several sessions have focused on the outcome of this investment into innovation. I put together a list of products and solutions unveiled over the last couple of years or are being worked on as we speak.
These are just names that I can remember, and given the inefficient RAM technology that my brain was shipped with, I am sure to have left out a few. The line-up is awe-inspiring, and the next 18 months are actually going to see an acceleration in the introduction of game-changing offerings.
|
Systems and Components |
Software |
Storage |
Development/Deployment Models and Services |
|---|---|---|---|
|
Sun Fire X4500 (Thumper) |
|||
|
Sun Storagetek 5800 (Honeycomb) |
|||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
Another thing that strikes one is the dilemma of which column to assign a technology to. The distinctions between server, storage, networking and software is blurring to a point where we can introduce terms like chipvers, storvers and servage, as one of our keynote speakers Jim Baty pointed out more than a year ago.
Tags : suncec2007
Painting Vegas Red (or The Greening of Vegas)
Four thousand Sun engineers invaded the Paris and Bally's Hotels in Las Vegas earlier this week. Engineers usually act on the basis of logical deduction, a faculty that has seemingly limited application on the Las Vegas Strip. However, our collective presence (for the annual Sun Customer Engineering Conference) has not influenced the rampant speculation in the casinos one whit.
The northernmost hotel on the Las Vegas Strip is the Stratosphere. After a couple of days at the general sessions, breakout sessions and the UltraSPARC T2 based systems launch, i feel like I have been launched into the stratosphere. The heady feeling that one gets when surrounded by fellow professionals from around the world is something that will last for a while, I suspect.
The theme of the conference is Shift : Our Universe, Our World, Your Move. A key trend influencing Sun's technology directions is what we believe to be an imminent shift in the applications that will fuel demand for compute infrastructure. Application areas that our customers regard as their competitive weapons and that will scale massively will outstrip workloads that are cost centres for customers, in their demand for systems, storage, software and associated services. We call such application areas Redshift applications.
Another shift that our world is experiencing, and IT deployments particularly so, is the shift to green - ecologically responsible ways of working and living make economic sense as well as ease the strain on finite natural resources.
These are the shifts that our conference dwells on, and every one of us will be pondering what move (s)he needs to make to tailor our solutions and technologies to a Redshifted, Greener world.
Tags : suncec2007
Monday Aug 27, 2007
It is a classic contrarian approach : the best way to allow a chip to communicate with another might be to remove the wires that connect them. Chips today communicate with other chips through wires that are generally welded to the chip through a process called Ball Bonding. There are several challenges with this, and the situation gets exacerbated by the kind of performance and compute density improvements happening within the chip.
Chip to Chip communication is far slower than that within a chip (primarily because the width of, and spacing between, wires within the chip are a 100 times smaller than those of wires between chips. This obviously is a performance bottleneck. Besides, an elaborate process is required to wire a chip up, making it expensive.
Proximity Communication lines up a transmitter on one chip against a receiver on another chip, with the two chips positioned very close to each other. The two form a capacitor and a voltage driven through the transmitter will result in a corresponding charge on the receiver, thus achieving communication. The method has been shown to achieve an I/O pad density 60 times greater than that possible with Ball Bonding. It thus allows larger bandwidths for a given area on the chip.
Other advantageous side-effects follow too : the conventional wire communication needs signal amplification as the external wires are larger. The power consumed and heat dissipated also increases because of the energy needed. Proximity Communication obviates the need for high amplification, resulting in smaller transmit/receive circuits, and expensive cooling. As the chips are not wired up to each other, the technique also permits chips to be easily replaced. One of the first patents in the area was awarded to Dr. Ivan Sutherland, Sun Fellow and Vice President. The idea was the overall Gold winner in The Wall Street Journal's Technology Innovation Awards in 2004.
The technology caused a stir at the Sun Labs Open House in April 2007. Dr. Hans Eberle is working on Project Sedna, a next generation datacenter switch which employs Proximity Communication.
Key to the practical deployment of proximity communication is solving the problem of rotational and translational mis-alignment : if the transmitters on one chip do not align perfectly with the receivers on the other, the coupling might not be strong enough for communication to take place. To deal with this, Greg Papadopoulos and Robert Bosnyak developed Electronic Alignment, a technique that compensates for mis-alignment by allowing transmit/receive pad positions to move correspondingly. Dr. Drost credits this technique as vital to making the technology practical.
An amusing sidelight is provided by the title of the illustration in The Register's Proximity Communication article : Chip Fornication. Considering that the chips are not in contact with each other, El Reg must have meant Coupling 2.0.
Real Time Matrix and the SF T2000 : A Perfect Match
ServerWatch has an article that details how the UltraSPARC T1 based SF T2000 powers the Real Time Matrix's ability to process millions of items a day and generate matches against millions of preferences.
Real Time Matrix provides technology to find what is relevant to a user from the flow of live Web content and sends it to a connected device of your choice. The iJ.am website is an example of how the technology works.
Quotes from the article :
We need a stable, robust infrastructure to process millions of items a day, match against millions of preferences and run 24/7. For high-speed, high-performance, 100 percent raw computing, we are finding it is cheaper and better on Sun and Solaris
When Real Time Matrix replaced a dual socket dual core system running Fedora Core with the SF T2000 running Solaris, they found that their matching throughput went up from hundreds per second to 10000 per second.
Although (the UltraSPARC T1) has up to 32 threads, Whitehead said it currently uses only up to 20 threads to achieve the 10,000 match per second rate
Tags : SPARC Solaris CoolThreads Niagara
Monday Aug 20, 2007
Tilera has announced availability of the TILE64, a chip with 64 cores (tiles) interconnected by a mesh-like on-chip network. Targeted at advanced networking and multimedia processing, the chip features 64 tiles each with a 3 VLIW pipeline processor, L1 and L2 caches, as well as a non-blocking switch that meshes the tile with the others on the chip. The chip provides the option of running an independent OS on each tile or an OS spanning multiple tiles.
A c|net News article has more details on the chip. The Tilera website also has an architecture brief.
Dr. Agarwal, CTO and co-founder of Tilera, has earlier worked on the MIPS processor used notably by SGI, and is leading the RAW project at the Computer Science and Artificial Intelligence Laboratory, MIT.
Rock and Transactional Memory
Applications have always striven for parallelism to take advantage of multiple CPUs. One of the things a program has to take care of in such an environment is data that is concurrently read or written by different threads or jobs. Traditionally this is done through Locks. A variety of locking techniques are used to synchronize data access, but generally problems encountered by programmers include complexity and overheads if the locks are too fine-grained, bottlenecks and lack of scaling if they are too coarse-grained, and concurrency issues introduced by the locks themselves. The problem will escalate as even single sockets now contain multiple cores, each of which can handle multiple threads.
Enter Transactional Memory. The concept is similar to transactions conducted on a database. Database transactions usually have ACID (Atomicity, Consistency, Isolation and Durability) properties that ensure correct processing and data integrity. Transferring money from your account to another, for instance, might involve several database operations but the latter are treated as one atomic transaction and either succeed altogether or fail altogether, thus ensuring no intermediate state. The same concept is being adapted for data operations in memory, in an effort to obviate the need for a programmer to worry about interactions between concurrent threads or operations.
Transactional Memory can be implemented in software or in hardware or as a hybrid of the two. In a Spotlight Article at Sun Labs, the research being carried out at the labs on transactional memory is detailed. Apart from research details, the article make note of the support in Rock for hardware transactional memory, the first ever implementation in a general purpose processor.
InternetNews.Com has an article discussing the announcement. Marc Tremblay discussed transactional memory support in his keynote at PODC last week.
Tags : Rock SPARC Concurrent Programming
Monday Aug 13, 2007
I took this photograph on our way to Amby Valley, for the Sun India Kickoff, struck by the (unintentional) relevance of the phrase N2 Pressure. The UltraSPARC T2 (codenamed Niagara 2) launch was just a couple of weeks away at the time, and it reminded me of how Throughput Computing applied enormous pressure on other general purpose processor manufacturers to basically change their design approaches.
The UltraSPARC T2 is set to further turn the screws on rivals. The specifications of the 342 square millimetre chip are mind boggling :
Full hardware support for 64 strands or threads
16 integer execution pipelines
08 floating point execution pipelines (with full support for the UltraSPARC VIS extensions)
08 memory pipelines
128KB Instruction Cache (16KB/Core)
64KB Data Cache (8KB/Core)
A Core x L2 Cache crossbar (180GB/s read and 90GB/s write, meaning we would make iTunes load claims too, if we didn't blush when making them)
04 MB L2 Cache
04 memory controllers
08 cryptographic co-processing units
An 8 Lane PCI-Ex Controller
02 1/10 Gigabit Ethernet Controllers
Like the French are wont to say, Le microprocesseur est mort, Vive le microsystem! All of this consumes less than 95 Watts at peak when operating at 1.4GHz. Less than 1.5 Watts per thread.
The combination of a system with the chip, Logical Domains and Solaris Containers make for a compelling virtualization platform to run application workloads that are thread rich. We are told, by IBM and by HP (surprise! surprise!), that this will be applicable only in some niche markets. I expect to hear soon that they will be exiting the transactional processing niche market and the network computing niche market and the ERP niche market and the ...
HP also puzzlingly called the Niagara platform a proprietary architecture. On second thoughts, since they feel that the Internet market is a niche, the maker of industry standard platforms might not have come across the OpenSPARC website. Jokes apart, the UltraSPARC T1 based systems were the fastest ramping products in Sun's history, and that might be only the first step for Niagara-kind.
Small wonder that when John Hennessy and David Patterson were putting together the fourth edition of the classic Computer Architecture : A Quantitative Approach, the UltraSPARC T1 was one of two contemporary processors (The AMD Opteron was the other) to feature in the case studies.
Wednesday Jun 20, 2007
The word, of course, means divination by numbers. The Wikipedia, the fount of all wisdom, tells us that divination is more formal or ritual in character, as opposed to fortune-telling. I guess the unwashed masses resorted to fortune-telling, while royalty deigned to consult bird entrails or numbers, as the case may be. It is also spelt (and presumably pronounced) Arithmomancy. Perhaps numerology prompted the alternative spelling.
What has this got to do with processors? The IBM Benchmarking and Systems Performance blog makes the shocking assertion that 4.7 > 1.4, in a reference to the clock speeds of the POWER6 and UltraSPARC T1 respectively. The post is a response to BM Seer blogging about how the single socket System p 570 SPECjbb2005 result has been beaten by Dell PowerEdge 860 and Sun Blade T6300 single socket results.
Choosing to focus on the word faster, and interpreting it literally, is a touch disingenuous, I think : we are all in agreement that the POWER6 designers did a fantastic job in achieving a speed of 4.7GHz. However, like John Meyer explains so lucidly, the achievement is like winning a race declared pointless half-way through. If 4.7GHz is to signify anything vis a vis 1.4GHz, the frequency difference should translate into compelling system price/performance metrics. Unfortunately, this doesn't seem to be the case. The SPECjbb2005 results[1] show that a socket-full of 1.4GHz UltraSPARC T1 beats, well, a socket-full of 4.7GHz POWER6. SPECjbb2005 results do not include price disclosures.
I don't know how Arithmancy deals with real numbers - pesky things like 4.7 and 1.4 that have decimal places. However, I came across this Arithmancy Calculator at The Sorcerer's Companion, and having nothing more productive to do at 4:30AM, plugged in the words POWER6 and Niagara. Here is what came up, for those of you wanting to save a mouse click and some typing :
POWER6
Character Number : 5
Five is the number of instability and imbalance, indicating change and uncertainty. Fives are drawn to many things at once but commit to none. They are adventurous, energetic and willing to take risks. They enjoy travel and meeting new people but may not stay in one place very long. Fives can be conceited, irresponsible, quick-tempered and impatient.
Niagara
Character Number : 6
Six represents harmony, friendship, and family life. Sixes are loyal, reliable, and loving. They adapt easily. They do well in teaching and the arts, but are often unsuccessful in business. They are sometimes prone to gossip and complacency.
Hmm. I can understand the proclivity to gossip : who wouldn't when there is so much to boast about? I don't know about the lack of business success and complacency though : UltraSPARC T1 based systems went past the half billion dollar revenue mark in news (slide 6) from the latest quarterly results, and Niagara 2 based systems will be shipping later this year. Design objectives of the Niagara 2 are to double throughput in the same power and thermal envelope. 64 threads, 8 FPUs, on-chip networking, enhanced encryption/decryption capabilities ... if Julius Caesar occupied a socket, he would have said "yond Niagara 2 has a lean and hungry look; it throughputs too much: such chips are dangerous".
[1] SPECjbb2005 Disclosure Statement :
Sun Fire T6300 (1 chip, 8 cores) - 96523 SPECjbb2005 bops, 24131 SPECjbb2005 bops/JVM
IBM System p 570 (1 chip, 2 cores) - 88089 SPECjbb2005 bops, 88089 SPECjbb2005 bops/JVM