The trend in HPC is moving towards higher productivity
INTERVIEWAn interview on HPC trends and on the development of transtec with Dr. Oliver Tennert, Director Marketing & HPC Solutions at transtec
To date, transtec has installed over 400 clusters, with renowned research institutes in its list of customers. Dr. Tennert, how would you describe transtec's position?
Tennert: For a long time High Performance Computing was an area where size was the central category, for technological and financial reasons. It was quite logical that the big players were at the centre of attention. With this, I don't just mean the supplier's side with names like IBM or HP but also the users. Until recently, HPC was the domain of large industrial or scientific research institutes. While extending our HPC activities we have noticed that the situation is changing quite strongly. The market for HPC solutions for medium-sized enterprises is increasingly starting to move. Medium-sized enterprises are also being discovered as customers by the big companies and, apart from that, more solutions are being developed which are suitable for medium-sized enterprises. We find that we are in an ideal position here, as transtec has 30 years of experience in the development of IT solutions for science & research applications. Alone the 400 clusters that we have installed speak for this as do our long-term co-operations with partners such as the Dresden-Rossendorf Research Centre or the Technical University in Darmstadt. In contrast to many other players in the market, we also have 30 years of experience in the design, implementation and care of IT solutions for medium-sized enterprises. We know what counts there and that is not primarily size but cost-effectiveness and efficiency. These are criteria which a medium-sized enterprise views quite differently to a large research institute. Our technology partners in the HPC industry have acknowledged this. There is a reason why transtec is a distributor for such renowned names as Adaptive Computing and CRAY.
Let's stay with Adaptive Computing. What is it that makes a product like their Moab Cluster Suite so interesting for transtec? Where do you see the particular advantages of this system compared to other solutions?
Tennert: Operating costs are an important criterion when we discuss setting up an HPC cluster with our customers. The working time and resources required to manage the cluster is an important point there. With the Moab Cluster Suite, Adaptive Computing has developed a management tool which addresses exactly this point. The Moab Cluster Suite family, whichever member is running on the cluster, provides an individual high level management system. Using the graphical user interface, apart from experienced administrators, users too can monitor and even control their own jobs. It is not one of those complicated script-based solutions which are found so frequently in the HPC area. Let's take the Moab Adaptive HPC Suite as an example. This is capable of altering the operating system for a computing node during normal operation, according to the requirements of the current workload, without anyone having to initiate or monitor the process.
At this year's ISC in Hamburg, a very special agreement was completed. transtec and CRAY have entered into a long term partnership. The HPC specialist from Tübingen in southern Germany is thus the only large German supplier in Europe to offer the desk-side CX1 supercomputers from CRAY, the market and technology leader from the USA.
Tennert: This agreement shows that we have done everything right. CRAY is absolutely the brand for supercomputing. We view this partnership under two aspects. On the one hand it completely subscribes to our brand focus on HPC solutions, which we will immediately further extend and strengthen. Here, working together with CRAY is simply magnificent. On the other hand, this also supports our aim to be the company which drives along the transmission of high end technologies into the working conditions of medium-sized enterprises. With CRAY CX1 systems, High Performance Computing thus moves into the office area.
Bill Dally, Stanford-Professor and chief scientist at Nvidia, recently prophesied to Forbes magazine a computer world in which parallel systems permanently lead to increases in performance. If he has anything to do with it, the solution is called: "Processors with more parallelism", in other words, computing units which can calculate tasks simultaneously. Do you agree with his opinion?
Tennert: Yes, absolutely. Even in the 1990s, a start was made to improve performance via simple parallelised procedures. Particularly the parallel aspects of course increase the performance, but they also guarantee much better scalability of the whole system. The requirements today are for a system which runs without a lot of maintenance while being easy to modify. Even in the area of storage, some computers exploit parallelism. In comparison to scale up solutions, parallel systems allow significant acceleration of computing power to be achieved for significantly lower investment costs.
At this year's ISC in Hamburg, Kirk Skaugen from Intel will speak about scale-up and scale-out in supercomputing. You will be speaking at ISC too. What will your subject be?
Tennert: On 2nd June at 3:30pm, I will be participating in a "Hot Seat" session. The subject is "30 Years of Experience in Scientific Computing – Performance Transforms into Productivity".
That fits. In this context you often speak about the double meaning of the "P" in "HPC": "Performance" plus "Productivity". Is there more to that than a marketing slogan?
Tennert: At transtec, for HPC, we translate "Performance" into "Productivity" because this is what our customers think. For most of them, it is not the pure performance of the hardware which we can install that counts. Especially medium-sized enterprises are looking for the solution to an IT problem and they expect that their productivity will be increased by that. We don't sell a ton of metal, but an individual solution for more productivity. At company X quite different requirements have to be considered here than for company Y. One quite central factor is the ease of use.
Even experienced IT administrators or large IT departments place increasing value on restricting the costs of administering clusters, so that the complexity of the management does not quickly consume the efficiency increases resulting from the computing power. For smaller companies this applies even more clearly. The time and know how are often simply not available.
Let's risk a look at the future of supercomputing. Many experts now only speak of the cloud. Large cloud providers like Amazon or Google now rely on the structures. Will classic supercomputers go out of fashion in the age of cloud computing?
Tennert: No, I don't think so. Cloud computing is still in its infancy and, in my opinion, cannot displace the local systems, even in future. For example, the cloud is much too dependent on the speed of the networks. However, exactly that is one of the most important keys in supercomputing. If the flow of data is once held up, or if the system cannot get the data out of memory as fast as is possible, then cloud computing quickly shows its limits. If a supercomputing system is quickly scalable and if it parallelises correctly, then customers will be very happy with the way that jobs are processed. transtec is thus quite relaxed about the hype surrounding cloud computing. We are also supported here by the pure sales figures: Medium-sized enterprises continue to invest in clusters and hardware and they are not simply going to give up on their investment. I think that cloud enthusiasts underestimate the thinking and functional logic of medium-sized enterprises: To release critical company data and processes completely to the cloud and thus to another company is anathema to many of our customers' sense of security. Not to mention the legal uncertainty. I presume that cloud computing will first find its market with users who work in collaboration, but who do not wish to use any resources to build up the necessary architecture – and of course for private users to whom the large Internet service suppliers increasingly offer cloud computing solutions as "apps". For an average company I do not see that happening so far. And I say that as someone who bears responsibility for a medium-sized enterprise.
This press release and a photo of Dr. Oliver Tennert can be downloaded at: http://www.panama-pr.de/download/transtec-InterviewOliverTennert.zip
Tennert: For a long time High Performance Computing was an area where size was the central category, for technological and financial reasons. It was quite logical that the big players were at the centre of attention. With this, I don't just mean the supplier's side with names like IBM or HP but also the users. Until recently, HPC was the domain of large industrial or scientific research institutes. While extending our HPC activities we have noticed that the situation is changing quite strongly. The market for HPC solutions for medium-sized enterprises is increasingly starting to move. Medium-sized enterprises are also being discovered as customers by the big companies and, apart from that, more solutions are being developed which are suitable for medium-sized enterprises. We find that we are in an ideal position here, as transtec has 30 years of experience in the development of IT solutions for science & research applications. Alone the 400 clusters that we have installed speak for this as do our long-term co-operations with partners such as the Dresden-Rossendorf Research Centre or the Technical University in Darmstadt. In contrast to many other players in the market, we also have 30 years of experience in the design, implementation and care of IT solutions for medium-sized enterprises. We know what counts there and that is not primarily size but cost-effectiveness and efficiency. These are criteria which a medium-sized enterprise views quite differently to a large research institute. Our technology partners in the HPC industry have acknowledged this. There is a reason why transtec is a distributor for such renowned names as Adaptive Computing and CRAY.
Let's stay with Adaptive Computing. What is it that makes a product like their Moab Cluster Suite so interesting for transtec? Where do you see the particular advantages of this system compared to other solutions?
Tennert: Operating costs are an important criterion when we discuss setting up an HPC cluster with our customers. The working time and resources required to manage the cluster is an important point there. With the Moab Cluster Suite, Adaptive Computing has developed a management tool which addresses exactly this point. The Moab Cluster Suite family, whichever member is running on the cluster, provides an individual high level management system. Using the graphical user interface, apart from experienced administrators, users too can monitor and even control their own jobs. It is not one of those complicated script-based solutions which are found so frequently in the HPC area. Let's take the Moab Adaptive HPC Suite as an example. This is capable of altering the operating system for a computing node during normal operation, according to the requirements of the current workload, without anyone having to initiate or monitor the process.
At this year's ISC in Hamburg, a very special agreement was completed. transtec and CRAY have entered into a long term partnership. The HPC specialist from Tübingen in southern Germany is thus the only large German supplier in Europe to offer the desk-side CX1 supercomputers from CRAY, the market and technology leader from the USA.
Tennert: This agreement shows that we have done everything right. CRAY is absolutely the brand for supercomputing. We view this partnership under two aspects. On the one hand it completely subscribes to our brand focus on HPC solutions, which we will immediately further extend and strengthen. Here, working together with CRAY is simply magnificent. On the other hand, this also supports our aim to be the company which drives along the transmission of high end technologies into the working conditions of medium-sized enterprises. With CRAY CX1 systems, High Performance Computing thus moves into the office area.
Bill Dally, Stanford-Professor and chief scientist at Nvidia, recently prophesied to Forbes magazine a computer world in which parallel systems permanently lead to increases in performance. If he has anything to do with it, the solution is called: "Processors with more parallelism", in other words, computing units which can calculate tasks simultaneously. Do you agree with his opinion?
Tennert: Yes, absolutely. Even in the 1990s, a start was made to improve performance via simple parallelised procedures. Particularly the parallel aspects of course increase the performance, but they also guarantee much better scalability of the whole system. The requirements today are for a system which runs without a lot of maintenance while being easy to modify. Even in the area of storage, some computers exploit parallelism. In comparison to scale up solutions, parallel systems allow significant acceleration of computing power to be achieved for significantly lower investment costs.
At this year's ISC in Hamburg, Kirk Skaugen from Intel will speak about scale-up and scale-out in supercomputing. You will be speaking at ISC too. What will your subject be?
Tennert: On 2nd June at 3:30pm, I will be participating in a "Hot Seat" session. The subject is "30 Years of Experience in Scientific Computing – Performance Transforms into Productivity".
That fits. In this context you often speak about the double meaning of the "P" in "HPC": "Performance" plus "Productivity". Is there more to that than a marketing slogan?
Tennert: At transtec, for HPC, we translate "Performance" into "Productivity" because this is what our customers think. For most of them, it is not the pure performance of the hardware which we can install that counts. Especially medium-sized enterprises are looking for the solution to an IT problem and they expect that their productivity will be increased by that. We don't sell a ton of metal, but an individual solution for more productivity. At company X quite different requirements have to be considered here than for company Y. One quite central factor is the ease of use.
Even experienced IT administrators or large IT departments place increasing value on restricting the costs of administering clusters, so that the complexity of the management does not quickly consume the efficiency increases resulting from the computing power. For smaller companies this applies even more clearly. The time and know how are often simply not available.
Let's risk a look at the future of supercomputing. Many experts now only speak of the cloud. Large cloud providers like Amazon or Google now rely on the structures. Will classic supercomputers go out of fashion in the age of cloud computing?
Tennert: No, I don't think so. Cloud computing is still in its infancy and, in my opinion, cannot displace the local systems, even in future. For example, the cloud is much too dependent on the speed of the networks. However, exactly that is one of the most important keys in supercomputing. If the flow of data is once held up, or if the system cannot get the data out of memory as fast as is possible, then cloud computing quickly shows its limits. If a supercomputing system is quickly scalable and if it parallelises correctly, then customers will be very happy with the way that jobs are processed. transtec is thus quite relaxed about the hype surrounding cloud computing. We are also supported here by the pure sales figures: Medium-sized enterprises continue to invest in clusters and hardware and they are not simply going to give up on their investment. I think that cloud enthusiasts underestimate the thinking and functional logic of medium-sized enterprises: To release critical company data and processes completely to the cloud and thus to another company is anathema to many of our customers' sense of security. Not to mention the legal uncertainty. I presume that cloud computing will first find its market with users who work in collaboration, but who do not wish to use any resources to build up the necessary architecture – and of course for private users to whom the large Internet service suppliers increasingly offer cloud computing solutions as "apps". For an average company I do not see that happening so far. And I say that as someone who bears responsibility for a medium-sized enterprise.
This press release and a photo of Dr. Oliver Tennert can be downloaded at: http://www.panama-pr.de/download/transtec-InterviewOliverTennert.zip