Posts tagged with GigaOm
Thunder Clouds
Posted by Margaret Lewis in 2:20 PM
(2 votes, average: 4.50 out of 5)
Loading ...In my mind a good conference is one that generates controversy – with my roots firmly in the 60’s I’d take revolution over evolution any day. The GigaOM Network’s Structure09 “Putting Cloud Computing to Work” delivered in my book as a thought-provoking conference.
AMD, in collaboration with HP, were Primetime Sponsors of Structure09, which drew the “who’s who” of cloud computing – hardware, software, providers, consulting, press and users. One of the featured sessions was a Fireside Chat featuring Jonathan Heiliger, vice president of technical operations at Facebook, and Om Malik, founder of the GigaOM Network.
In this session Jonathan expressed his disappointment with AMD, Intel, and system OEMs. Specifically for AMD and Intel, Jonathan lamented that he was not getting the advertised performance from our latest processor technology. On the system provider front, Jonathan’s unhappiness was focused on the power consumption of the system.
As a representative of one of the companies called out by Jonathan – and sitting in the audience and hearing his talk first hand – it was hard not to squirm in my seat. We are supposed to be customer centric – we are supposed to be solving Jonathan’s problems so he can successfully grow his business.
So what is the take away from Jonathan’s talk? Two points: benchmarks don’t necessarily measure real world performance, and the harsh reality of living on the “bleeding edge.”
Benchmarks
The first point – benchmarks – is a topic that we at AMD are talking about a lot. Be it server VMmark scores or laptop battery life measurements – we want performance measurements more relevant to real-world situations. Jonathan’s comments validate this approach – he has a very specialized software stack that he believes is not showing the performance of the industry standard server benchmarks that AMD and Intel quote.
Life on the Bleeding Edge
The second point is life on the bleeding edge – Jonathan’s words suggest he is trying to drive his production environment to new levels of scalability. When you hear Jonathan’s pain you realize why the research community and supercomputing world are often the leaders in implementing new innovations – they have the luxury of doing this type of revolution in non-production environments.
Perhaps Facebook has found the perfect storm. Getting the most value out of the fastest processors in the world is hinged on hardware and software working in lock step, and innovating together. This is an issue we deal with every day in developing and optimizing our multi-core technology. But, as an industry, we still need to get better.
So, how are we addressing this issue? At AMD we are focusing our efforts on designing and delivering balanced server platforms in terms of performance, price, and power consumption. The market trends of virtualization and cloud computing confirm this approach – these are not “high performance computing” workloads where raw processor power is demanded. Rather, these hard working environments need to keep utilization rates high and user response time low – all while drawing as little power as possible.
In terms of driving software to deliver better multi-threaded code, AMD is investing resources to design better optimized compilers and working to develop technologies that help with the development of parallel programs, like OpenCL.
There is a revolution coming – the way we use technology is changing. Jonathan’s talk was a reminder to all of us who design hardware and software: we need to stay connected with the customer. Designing technology for technology’s sake is not the answer.
What are your thoughts on this topic? I’d love to hear your take on it.
Margaret Lewis, director, AMD commercial software and solutions
Margaret Lewis (@margaretjlewis) is a Product Marketing Director at AMD.
Her postings are her own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
@margaretjlewis
Of Cloud and High Performance Computing Clusters
Posted by Margaret Lewis in 4:05 PM
The Summer Solstice kicked off an interesting week featuring two computing conferences that at first glance seem to be at opposite ends of the spectrum. From June 23 to June 26: the International Supercomputing Conference in Hamburg, with a focus on bleeding edge, high performance computing (HPC). June 25: the GigaOM Network presents a premiere event for Cloud Computing and Internet infrastructure, Structure09 in San Francisco (I’ll post more on my thoughts from this conference tomorrow).
Now you might ask, “what do these events have in common?” The answer ─the heart and soul of both HPC and Cloud Computing is the “compute cluster” -an approach that links groups of computers together and has been driven into mainstream by x86 computers. And while the applications these clusters are driving and the data they are delivering might be drastically different – there are some basic commonalities on an infrastructure level.
Both types of clusters thrive on multi-core processors. More cores typically deliver more compute capabilities that can translate into performing more calculations for the HPC world and handling more transactions for the Cloud world. Power consumption is a nagging problem since larger and more powerful clusters commonly eat up a lot of energy and demand a lot of cooling. Both are driving new advancements in our digital world: HPC clusters help us solve some of today’s most complex problems while cloud computing provides the framework for searching and sharing the answers. And, of course, there is this bothersome economy influencing how many servers can be added to any type of data center.
So, what is AMD doing to design for both types of clusters? We continue to drive a balance of price, performance, and power into our processor architectures. Our server processor road map demonstrates an understanding of the real needs of both HPC and cloud clusters. Our G34 platforms, under the platform name of “Maranello,” are designed for expandability and performance. Our C32 platforms, under the platform name “San Marino,” help provide the low cost and power consumption that many cloud providers require.
I spent time last month in AMD’s Bellevue office and met with a talented group of AMD engineers who are working on further developing their understanding of the hardware and software requirements for cloud computing so they can further optimize AMD processor-based platforms. As a company we are continuing our efforts to help refine the underlying technology that will enable both types of clusters in the future─toward the goal of improved platform power efficiency, software methodologies to efficiently drive heterogeneous cores, enhanced graphics capabilities for improved user experience, and the continued maturation of virtualization for x86-based computers.
My colleague John Fruehe published a blog that takes a closer look at why he believes the AMD OpteronTM processor is ideal for supercomputing. I will similarly be posting a blog on cloud computing and giving you my first hand view of the Structure09 conference.
I can’t resist ending on a HPC note – particularly since, in a galaxy far away (it was actually last century), I was a part of the HPC community, working at the Maui High Performance Computing Center. Take a close look at the Top500 list and notice that AMD Opteron processors are in the #1 and #2 supercomputers – as are 9 of the top 20 computers. At AMD we design for innovation and quality!
Let me know your thoughts – what similarities do you see between HPC and Cloud clusters?
Margaret Lewis, director, AMD commercial software and solutions
Margaret Lewis (@margaretjlewis) is a Product Marketing Director at AMD. Her postings are her own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
@margaretjlewis