Pawsey SC15 Blog Entry #3 – Karina Nunez
This year SC15 broke attendance records, with more than 12,903 registered attendees on Thursday November 19 alone, while the exhibit hall featured 343 exhibitors from industry, academia and research organisations*.
During the three days of exhibition, the Australian HPC booth was approached by more than 500 people wanting to find out what is happening in Australia in terms of HPC capabilities, asking questions about the SKA and Pawsey, what industry engagement models are used, how Australia is connected with other supercomputing facilities…and they all took a piece of Australia back with them. We handed out 250 Pawsey wombats, 300 TimTams and more than 500 Australian HPC lizards, along with extensive information about our capabilities.
We also took the opportunity to let people know about the Head of Supercomputing position that Pawsey is currently advertising.
Many of the TimTams were enjoyed during the popular Australian HPC TimTam talks, held at lunch time every day, with this year’s presentations focusing on the Square Kilometre Array, bioinformatics research, projects of Australian national significance and storage and cloud computing, all delivered by Pawsey and VLSCI staff members.
On Wednesday November 18, Pawsey supercomputing application specialist Charlene Yang presented her poster, alongside more than 130 others who were chosen to be part of this year’s Regular & ACM Student Research Competition poster session. She spent more than an hour answering questions about her topic “Adapting genome-wide association workflows for HPC processing at Pawsey” to SC15 attendees.
The presentation was well received and Charlene was invited to submit chapters for a book about how HPC transforms big data analysis.
People who came to the poster session were mainly researchers or working as the HPC support in a bioinformatics research group. From their responses it was clear many of the issues we see at Pawsey are universally shared, for example, how bioinformatic applications don’t run in parallel and how researchers struggle to construct a complicated pipeline that runs, and Charlene was able to discuss areas of Pawsey best practice. To check Charlene’s poster click here.
The Student Cluster Competition kept us all awake waiting to see the Pawsey team performance but it was won by Team Diablo, a team of undergraduate students from Tsinghua University in China. A team from Technical University of Munich, Germany, were also awarded, for achieving the highest performance for the Linpack benchmark.
The teams were judged base on the combined score for correctly completed workload, benchmark performance, demonstrated understanding of architecture and performance through profiling and analysis, and interviews.
The competition was a two day non-stop challenge, with each team partnering with vendors to assemble a small cluster with a 3,120-watt power limit, and working with application experts to tune and run competition codes. They competed to complete a real-world workload across a series of scientific applications. More details about the outcome of SCC15 can be found here.
Australia, represented by Pawsey, has been part of this competition for the last three years. Our team was made up of six undergraduate students from Murdoch University, all computer science majors. Each are focused on different aspects of computing, including game design, mobile app development, system administration, and cyber-forensics (Click here to download the Team poster). This year we received support from Cray support for our cluster and a team from Altair was on the floor with the students. The team worked really hard and learned lots during the process. And since the competition has now been replicated in Europe, we will be competing in next years’ ISC16.
The last day of the event, made all of us feel proud. It was a workshop about HPC user support tools organised by Australian representatives from Pawsey Supercomputing Centre and VLSCI, together with people from Lawrence Livermore National Laboratory, USA.
The workshop was a platform for system administrators, user support team members, tool developers, policy makers and end users to discuss support issues, and we lead that conversation. The program had presentations from University of Tennessee, Argonne National Laboratory, Florida State University, Kong Abdullah University of Science and Technology, Texas Advance Computing Centre, NASA and University of Arizona, all sharing their experiences and best practices with the audience. For more information about the program please visit http://hust15.github.io/
It was a great pleasure representing Australia at this international event and it is a bigger pleasure working with Pawsey staff to do great things for Australian science!
*More figures about SC15 here.
Pawsey SC15 Blog Entry #2 – Chris Schlipalius
The week started out on a Sunday with technical briefings including the GPFS user group afternoon, in which case studies were provided. An IBM technical roadmap was outlined to attendees, and a request for user group input was sought by IBM on a number of planned enhancements for monitoring and displaying information on GPFS storage.
GPFS is the filesystem used on the RDS storage system we purchased from DDN. There are some good enhancements coming up for small files, of which many of our projects consist, so this is a good direction. A new GUI is available for Storage Administrators, and backups are improved – all wins for us next year potentially.
I made a number of site contacts (speakers from organisations in Asia and the UK) as well as with IBM, so I hope to keep those options open. Organisations were also interested in us providing a case study to them, as we have a very large amount of GPFS storage for any research organisation.
Next, I had a number of IBM workshops under non disclosure agreement (NDA), mainly talking about the finer detail on the new hardware side.
The following day was the parallel storage workshop, in which I learned about new developments in this sphere, including a 3D storage technology media pilot that might be linked to SpectraLogic. We could perhaps access this as it has a 20 year guaranteed media life for archival data which would be of particular interest for radio-astronomy data).
I had the DDN user group and meetings in which we were able to look at specifications for potential storage under Lustre. The DDN user group again featured some innovative and ‘problem solved’ type case studies. These were again good to hear, but many cases were on upgrades from very old storage hardware, and we run relatively new gear.
On the conference floor:
Wow, the biggest conference ever! I learned there were over 12,000 attendees!
SpectraLogic was very keen on doing a case study with us, as were Lockheed Martin on the use of LiveArc to power the Pawsey data portal (I spoke with their SysAdmin for over one hour and they were keen to have any papers from us on our use and setup).
The stand looked great, Karina was running it smoothly. People loved the wombats.
I tried to attend the stand as often as I could and fielded a lot of questions, often on our networking, our collaboration with other agencies and institutions, and internships!
I obtained many images of badges and many people were interested in being included on the Pawsey friends contact list
I held my presentation about “Leveraging Research Data Services for research of National significance at Pawsey Supercomputing Centre”, which I hope can be developed into something to share out more widely down the path.
I met with HGST who had no idea that real world statistics were being published on their drives (from Backblaze who run hundreds of thousands of commodity disk for cloud data customers), comparing them to WD and Seagate, and I explained that with our DDN storage, HGST underpins our statistics of less than four drive fails in nearly 12 months out of many hundreds of spinning disks.
I saw A-Star and a representatives of a US university who were keen to do some DTN service testing with us. Good if we can!
I touched base with NCI who are still keen to get their Livearc portal used, like we have with the Pawsey data portal, hoping to collaborate next year if it’s possible.
Lastly I spoke with the National Radio Astronomy Observatory and they were keen to hear about Lustre storage. I mentioned the DDN Lustre technology, and they were so interested in the potential value and stability and Intel support that they asked to go over to the DDN stand.
With a DDN pre-sale engineer (Goverment and Education) who spec’d out the option to suit, and NRAO were looking to get a (hopefully) good price on a buy later storage appliance to help with their data. I really hope they get a great deal, happy to push that for them too as they don’t have much money, do critical work and are wasting effort on old storage that has high overheads in staff maintenance effort and is also too small to suit. I was highly encouraged that DDN understood and would attempt to offer them the best price, great for colleagues all working in this area! If we can leverage our success (without cost to us) to enable better equipment and filesystems options for our colleagues in collaborating organisations, I feel we should do so if we can help.
So all in all, the conference was a big, interesting, friendly and informative experience for me. I enjoyed the personal-level interactions while on our stand as much as anything, plus I feel it’s a good time to be in big storage, stable and with more improvements coming, it’s not all rapid change to fix huge bugs, its more about tuning, and that’s best for us at this point. It’s what we in the storage technical team are looking for, stability and functionality enhancements.
Pawsey SC15 Blog Entry #1 – Karina Nunez
A year has passed since the Pawsey team went to New Orleans for the biggest meeting of all aspects of scientific and high-performance computing. This time we have come to Austin, Texas to find out what is new in the HPC world.
It is a really dense program, with lots to pick and choose from. This event keeps all its attendees – with more than 10000 international visitors every year – busy. There are workshops, tutorials, key notes, Birds of a Feather, poster, panel, and awards sessions (see the program at http://sc15.supercomputing.org/program).
The conference started on Sunday with a series of tutorials and workshops. However, the Pawsey Student Cluster Competition team has been working since Saturday, putting their Cray together and installing and running their system. It seems that they haven’t had much sleep, but spirits are still high. Let’s keep our finger crossed until Wednesday, when we will find out the outcome of the Competition.
This year the exhibition count has more than 350 exhibitors, with the Australian HPC booth among them. We are at booth #2633. This is the first time that Pawsey has been responsible for putting the space together, with representation from CSIRO, VLSCI and V3. The Pawsey team will be spending time at the booth to let people know more about Australia’s role in HPC, to discuss some of the most important projects running on our systems and to talk about the Australian way of doing things in this area. People attending the event are invited to adopt a wombat, or to put together a cardboard lizard, while they find out more about the different centres in Australia.
Monday night was the opening gala for the exhibition. The Australian HPC booth saw more than 100 visitors pass through and adopt a wombat, all wanting to know more about our systems characteristics, infrastructure, connexion speed, and more. All Pawsey staff were there to welcome visitors, with the space organised to allow meetings and to hold people for our popular Tim Tam talks – click here to download the Tim Tam talks program.
Yesterday the new top500 list was also released. Australia has five entries on this list. Pawsey dropped from 58th to 67th place, while VLSCI with Avoca placed 122nd and at entry 295 we find CSIRO’s GPU Cluster. There was also an industry system and NCI’s Raijin computer featured in the list.
Now we are looking forward to Tuesday’s poster session where Jindan (Charlene) Yang, Pawsey Supercomputing Applications Specialist, will be presenting about ADAPTING GENOME-WIDE ASSOCIATION WORKFLOWS FOR HPC PROCESSING AT PAWSEY (for the abstract click here) and for Thursday’s workshops, when Chris Bording, also a Supercomputing Applications Specialist, will present with some colleagues from Lawrence Livermore National Laboratory and VLSCI – HUST2015: Second International Workshop on HPC User Support Tools (more details here)
Stay tuned for more posts about SC15 in coming days!