Google's Urs Hölzle explains why beefier cores are better than whimpy cores

September 17, 2010 Dave Ohara

The Register covers a new paper by Google's Urs Hölzle.

Google ops czar condemns multi-core extremists

Sea of 'wimpy' cores will sink you

By Cade Metz in San Francisco • Get more from this author

Posted in Servers, 17th September 2010 07:04 GMT

Free whitepaper – The Reg Guide to Solutions for the Virtual Era

Google is the modern data poster-child for parallel computing. It's famous for splintering enormous calculations into tiny pieces that can then be processed across an epic network of machines. But when it comes to spreading workloads across multi-core processors, the company has called for a certain amount of restraint.

With a paper (PDF) soon to be published in IEEE Micro, the IEEE magazine of chip and silicon design, Google Senior Vice President of Operations Urs Hölzle – one of the brains overseeing the web giant's famous back-end – warns against the use of multi-core processors that take parallelization too far. Chips that spread workloads across more energy-efficient but slower cores, he says, may not be preferable to chips with faster but power-hungry cores.

The paper is here and only 2 pages long. And, when thinking what motivated Urs to write this paper i think it was his frustration that too many people are focusing on the number of cores to solve a problem and not taking into consideration what happens to the overall system when you try to solve problems with a bunch of whimpy cores vs. brawny cores.

We classify multicore systems as brawny-core systems, whose single-core performance is fairly high, or wimpy-core systems, whose single-core performance is low. The latter are more power efficient. Typically, CPU power decreases by approximately O(k2) when CPU frequency decreases by k, and decreasing DRAM access speeds with core speeds can save additional power.

Urs as usual uses excellent presentation skills to make his point in three areas.

First, the more threads handling a parallelized request, the larger the overall response time. Often all parallel tasks must finish before a request is completed, and thus the overall response time becomes the maximum response time of any subtask, and more subtasks will push further into the long tail of subtask response times. With 10 subtasks, a one-in-a-thousand chance of suboptimal process scheduling will affect 1 percent of requests (recall that the request time is the maximum of all subrequests), but with 1,000 subtasks it will affect virtually all requests.

In addition, a larger number of smaller systems can increase the overall cluster cost if fixed non-CPU costs can’t be scaled down accordingly. The cost of basic infrastructure (enclosures, cables, disks, power supplies, network ports, cables, and so on) must be shared across multiple wimpy-core servers, or these costs might offset any savings. More problematically, DRAM costs might increase if processes have a significant DRAM footprint that’s unrelated to throughput. For example, the kernel and system processes consume more aggregate memory, and applications can use memory-resident data structures (say, a dictionary mapping words to their synonyms) that might need to be loaded into memory on multiple wimpy-core machines instead of a single brawny-core machine.

Third, smaller servers can also lead to lower utilization. Consider the task of allocating a set of applications across a pool of servers as a bin-packing problem—each of the servers is a bin, and we try to fit as many applications as possible into each bin. Clearly that task is harder when the bins are small, because many applications might not completely fill a server and yet use too much of its CPU or RAM to allow a second application to coexist on the same server. Thus, larger bins (combined with resource containers or virtual machines to achieve performance isolation between individual applications) might offer a lower total cost to run a given workload.

How many data center operation VPs can write this paper? One. :-)

Keep the number of cores in mind for a green data center, smaller energy efficient processors may not be the most efficient overall.

Seagate and Samsung to co-develop SSD for enterprise storage

August 12, 2010 Dave Ohara

SSDs use much less power and have a higher performance under some conditions than HD, but the uptake in enterprises has been slow and in data centers that want to be green.

CNET has an article on Seagate and Samsung co-developing SSD for enterprise storage.

Seagate and Samsung to co-develop SSD controller

by Dong Ngo

Seagate and Samsung, the two major makers of hard drives and system memory, announced Thursday that they have entered into a joint development and licensing agreement.

Under this agreement, the two companies will develop and cross-license related controller technologies for solid state drives.

Seagate is leveraging its enterprise storage expertise.

Seagate says that the joint development will build on the existing SSD capabilities of each company while combining Seagate's enterprise storage technology with Samsung's 30 nanometer-class MLC NAND flash memory technology. Seagate will then use the jointly developed controller for its enterprise-class SSDs.

Seagate's blog says the partnerships is to address SSD memory errors and lower cost.

Each company brings something unique to the table besides its market leadership. While companies in any technology field when marketing will tend to focus on the positive aspects of the technology they are producing or selling, we know that behind the scenes all technologies have challenges and hurdles that must be overcome. In the case of storage, it doesn’t matter whether we’re discussing SSDs or HDDs; engineers working with both technologies are most often tasked with limiting the number of data errors produced at the media. Think of it as the game of always looking to make perfect something that will always be imperfect to start with. Seagate has great expertise in minimizing errors on its media and its current enterprise HDDs are best-in-class in the area of error recovery.

So that is at the heart of the collaboration from a technical perspective: error recovery and management. Samsung brings its flash technology expertise while Seagate brings its error recovery expertise to the table. Between them, the companies will look to produce a controller for SSDs that can attain the high levels of performance, reliability, and endurance demanded by enterprise storage applications.

Another interesting technical piece is the fact that today’s announcement references the use of Samsung’s 30 nanometer-class MLC (Multi-Level Cell) NAND as the technology base for the collaborative project. MLC NAND enables higher capacities at a lower cost, but it has not typically been a target technology for enterprise use due to having lower endurance. However, the controller technology that Seagate and Samsung develop together with its advanced error recovery and flash management, will enable more cost-effective and long-life products for the enterprise space.

Seagate's press release is here.

"Seagate has long recognized that solid state technology has an important role to play in the comprehensive solutions the storage industry will deliver today and in the future, particularly in the enterprise market," said Steve Luczo, Seagate chairman, president and CEO. "Today's agreement with Samsung will help us bring a compelling set of SSD innovations to the enterprise storage market, with benefits that range from enhanced performance, endurance and reliability to cost and capacity improvements. Overall, this agreement with Samsung strengthens our SSD solutions strategy, and positions Seagate well as global demand for storage continues on its strong growth path."

"We are pleased to be jointly developing a high-performance SSD controller with Seagate for the enterprise storage market," said Dr. Changhyun Kim, senior vice president and Samsung Fellow, Memory product planning & application engineering, Semiconductor Business, Samsung Electronics. "Our green memory solution is designed to enable more energy-efficient server applications, which is expected to increase the use of NAND-based SSD storage in enterprise applications."

As network speeds increase optical switching research increases with green and energy efficient benefits

July 8, 2010 Dave Ohara

Initially, I saw MIT's news announcing 100 times faster with less energy consumption, and dismissed the news.

An Internet 100 times as fast

A new network design that avoids the need to convert optical signals into electrical ones could boost capacity while reducing power consumption.

Larry Hardesty, MIT News Office

In today’s Internet, data traveling through optical fibers as beams of light have to be converted to electrical signals for processing. By dispensing with that conversion, a new network design could increase Internet speeds 100-fold.

June 28, 2010

The heart of the Internet is a network of high-capacity optical fibers that spans continents. But while optical signals transmit information much more efficiently than electrical signals, they’re harder to control. The routers that direct traffic on the Internet typically convert optical signals to electrical ones for processing, then convert them back for transmission, a process that consumes time and energy.
In recent years, however, a group of MIT researchers led by Vincent Chan, the Joan and Irwin Jacobs Professor of Electrical Engineering and Computer Science, has demonstrated a new way of organizing optical networks that, in most cases, would eliminate this inefficient conversion process. As a result, it could make the Internet 100 or even 1,000 times faster while actually reducing the amount of energy it consumes.

But, then here comes news from European Commission on Optical switching and green photonics.

C-3PO Strives for Green Photonics

JULY 8, 2010 | Craig Matsumoto | Post a comment

A European Commission-funded project is giving the optical components sector some green street cred by aiming to squeeze the power requirements for certain types of photonic devices.

The three-year project is being called Colorless and Coolerless Components for Low-Power Optical Networks, which participants are shortening to C-3PO.

Yes, sci-fi geeks, they named it after a breakfast cereal.

ADVA Optical Networking (Frankfurt: ADV) is announcing its involvement today, but C-3PO got introduced to the world a couple of weeks ago.

The author tries to tie in why energy efficient networks are important and the relationship with virtualization.

Photonics are not the biggest power suck in the network. But C-3PO participants reason thatevery piece of the network will eventually need a green checkup. "Every single layer in the complete network has to be considered, and that means network architectures, maybe even application architectures, systems, components, and safe or sleep modes. Optics can't be excluded," Grobe says.

The data center, as a whole, is a more obvious power hog, but Grobe points out that this problem is being addressed by virtualization, a process that lets users tap servers and storage that are widely distributed. Virtualization is made possible by high-speed optical networks. Thus, low-power optics will have some role in defining data-center power usage.

Intel Anthropologists Validates SeaMicro approach, multiple low power cores

July 3, 2010 Dave Ohara

ars technica has an interview with Intel CTO and discussed the role of anthropologists at Intel.

How Moore's Law drove Intel into the arms of anthropologists

By Jon Stokes | Last updated a day ago

MOUNTAIN VIEW — Intel CTO Justin Rattner took the stage at Intel's annual Research Day to host what was something of a launch party for Intel's new Interaction and Experience Research Lab—essentially a place to put all of the anthropologists and ethnographers that the company has been hiring over the past decade, and also a very high-profile validation of the value that the chipmaker places on the work of these folks.

The author wanted to dig into the impact of SeaMicro

Rattner also took up the topic of Atom vs. Xeon for cloud computing in a later Q&A session, and his response to my question of what he thought about SeaMicro's 512-Atom server might surprise you.

and here is what he uncovered.

"There's a growing body of evidence that suggests that for these massive datacenters there's a different optimum—a different set of tradeoffs—between performance and energy," he explained.

...

Rattner went on to say that the research backs up the idea that large clusters of fairly weak processors can be "dramatically more efficient" on certain types of cloud workloads than traditional enterprise servers (of the kind that Intel currently sells hardware for), which is why the company is eager to get to market with either SCCC or something like it.

iPhone 4 Antenna problems, world of analog in a digital space

June 30, 2010 Dave Ohara

I last wrote a blog post on June 17, 2010, and I have tons of ideas to discuss, but sometimes it is easier to get back into writing by starting on an esoteric subject.

iPhone 4 antenna problems.

Here is a post on Apple hiring 3 antenna engineers.

Three Apple job postings for iPhone / iPad antenna engineers to "Define and implement antenna system architecture to optimize the radiation performance for wireless portable devices." All three were posted on June 23rd, the same day that we started seeing widespread reporting of the left-handed reception issues. Coincidence?

and media is starting to cover Apple's discussion of known iPhone Antenna problems.

Leaked Docs Show Apple Knows About iPhone 4’s Flaws [REPORT]

According to documents leaked to Boy Genius Report today, AppleCare representatives are being given a strong company line to deliver to unhappy iPhone 4 owners who complain about reception issues.

Employees are told to say that the device’s reception performance “is the best we have ever shipped” and that itscritical antenna flaws are “a fact of life in the wireless world.” They are told not to perform service on iPhones with these problems and instead to give customers a PR-driven recitative instead.

In a nutshell, Apple knows the phone has problems but will insist that users are simply “holding it wrong.”

How can this happen?

Simple. Being an engineer at Apple working on analog technologies is not sexy.

When I worked at Apple (1985 - 1992), part of time I worked with a team where analog was a big part of the job - Macintosh II and Mac Portable - Power supplies and CRT Monitors. Some of these people had worked on the Apple II, original Mac and Lisa, and analog technologies were known as a must have skill to support the processing of bits. Thinking about signal waveforms, shielding, testing, trade-offs, and FCC Part 15 class B and A certification was part of every product development.

In a world of digital, few think of analog engineers. It is part of the communication problems between IT and the data center. Data Center people have analog experts. IT people can't see digital to analog issues.

But, there are a bunch of Apple people now learning about analog and antenna issues.

Can Apple make analog sexy? No, but they've helped increase the visibility that analog engineers are a critical part of a digital system.