Supercomputing booth/2014

From NorduGrid
Jump to: navigation, search

Details

Sc14-small.jpg
  • Web site
  • Dates: November 16-21, 2014
  • Location: New Orleans, LA
  • Booth nr: 3702
  • Booth personnel: Christian, Martin

Report

Participants

  • Christian, Martin

SC14 general notes

This year the convention was held at the Ernest N Morial convention center in New Orleans. This enormous center hosted another conference on public healt at the same time, and about a third of the center was empty. Although our booth was close to an entrance to the exhibition area, we were at the very far end of the convention center, more than a kilometer from the main entrance. We were also far from the big exhibitors. The only way to get from the technical sessions to the exhibition area was also at the very far end. This all meant that even though SC14 again broke size records, there was very little traffic passing our booth.

The Booth

The booth setup this year copied last years setup with the large poster covering the curtains and the roll up poster out by the aisle. No one complained about this, although it is technically prohibited. We also removed a siderail in the booth to create a more open look. With practice from last year set up was easily done in a couple of hours.

Materials Used

  • background poster printed on cloth (3.2 x 1.3 meter large). Plastic rails, hooks + long zip ties for adjustable hanging. Sides of poster taped to booth stand tubes.
  • A-REX roll-up
  • three chairs and the usual big table (provided by Freeman)
  • marketing cards similar to last year, but with an email adress added. 100 copies were printed and about 75 were left by the end of the show.

Budget

  • Booth fee: 400$
  • Personnel fee: 2 * 100 $
  • Travel cost(hotel+transportation): 2* ~17'000 DKK
  • Business cards printing: 33 $
  • Office supplies for booth building: <$10 (donated by Christian)

Aproximate booth cost: ~ 40'000 DKK

Log book

2014-11-16
We arrived at the convention center on Sunday 16, noticing that there was no carpet in our booth, only 3 chairs and a table was there. We contacted the freeman booth, where they first required a payment for a carpet. Insisting that we had ordered a carpet, and that our booth is a research one, and explaining that there was no way we could authorize payment without talking to people at home, they let us have a carpet for free.
We could not prepare the booth before the carpet arrived.
2014-11-17
Today, Monday, we realized we had no power in the booth. The power we used yesterday was for the booth behind ours - the cable is marked. It turns out we have not filled the form for ordering electricity. See [1] and the form [2]. Since we did not bring a projector we have no urgent needs for power. We talked to the booth behind us. They informed us they almost also did not get any power. There was no reminder to fill the power order, which is no longer part of the standard booth order form. There is no research booth power option on the power order form. Instead one should choose the option that is similar to what is included in the research booth and note on the form that this is for a research booth.
The booth selection process for next years SuperComputing (SC15) has changed. One can now wish a number of booth spaces. The booth space contract has to be submitting during SC14, in order to ensure a booth space for next year.
2014-11-18
In the morning we modified the business cards from last year adding an email address, and sending them to printing at the FedEx office in a nearby hotel.
Christian talked with Martin Swany from Indiana University who was researching HPC networks at Exascale. He would be happy to talk to us, if there was areas that would be of common interest in relation to Exercise.
we went to the European Exa-scale Projects booth, where we talked with a person on the Deep-er project, which is an ongoing hardware project, part of H2020. Most of the things we talked about, we had already learned from KnowARC and EMI.
In the afternoon we went to the Cray Europe Customer Session, where the Cray situation in Europe was presented. They talked about optimising energy consumption, for instance lowering the CPU load for tasks. They also talked about Exascale systems, and they presented an approach where a SSD layer is added between the node a storage, before hard disks. On the nodes they added a faster/more responsive memory layer between RAM and the CPU. On the storage side they showed a system where files/data automatically is moved from fast/expensive storage to slow/cheap storage based on parameters e.g. usage. In the end of the session we talked with Duncan who is chief for the European research project for Cray. He said they had cut the number of projects they were participating in, and only joining those they found interesting. We asked around for Alistair, but he had already left. Instead we talked to ? Alistair's boss, and head of EU projects at Cray. He told us that cray was consolidating their participation in EU projects. They were going to focus on about 3 projects that they believed were promising, among them were EXERSICE.
In the evening we went to the Cray reception where we finally met and talked with Alistair Hart about the EXERCISE project. Later on the chiefs chief of Miguel from CSCS joined the conversation, wanting to know what their/her role is in the EXERCISE project. Both of them sounded positive about the project.
2014-11-19
We picked up the business cards from the FedEx office. We had a visitor from the meterological office of the US navy. He did underwater weather presiction. He was mostly interested in talking physics.
2014-11-20
There were so few visitors to the booth we gave up sitting in it. Instead we armed ourselves with business cards and went around talking to prospects.
- Australien HPC (users)
We met Vera and chatted with her. We also attended a talk given by the leader of the Australian HPC center at Canberra NCI. We talked with him and a system administrator Robert Edberg from their
site. We discussed how they handled managing their 2k-3k users. It was all done by hand using old email templates etc. We discussed VOMS, and IGTF authentication authorization,
and how ARC is compliant with these. They were planning to look into a new system for managing users within the next 1/2 year, and were interested in using certificates and the
australian version of terena for user authentication. They wouldbe interested in talking to us. We exchanged business cards.
- Barcelona (simulation)
We asked around the exhibition if anyone had any experience with simulating performance, loads etc of cluster setups. Several people told us to talk to BCS who had some tool for :perfomance simulation.
At the BCS booth we talked to Judit Gimenez Lucas about simulation of clusters. She told us about their tool :Dimemas, which was mostly for MPI simulation, but might be interesting.
- SKA (Geant, Keith Grainge, Manchester)
We talked to Geant about possible grid users they were in contact with. They suggested we talk to Keith from the SKA. He was hard to track down, but we found him just as the
exhibition finished. We talked to him while NASA was dismantling their booth around us. He gave us an introduction to the SKA and how they are proceeding. There seem to be a
couple of factions within the SKA. Some who think it would be a good idea to build an infrastructure inspired what has been going on in the HEP community. Others who thought
that the astronomers would simply travel to the AU or ZA outback and bring the data back on their laptop. Keith was keen to continue discussing with us.


- ChinaGrid At the ChinaGrid booth they showed a poster with 10 or more sites connected by some means into a "grid". We tried to ask how they connect the sites and what software they used, but it was impossible to communicate.

- SLURM In the SLURM booth, a guy from NASAs Center for Climate Simulations (maybe Dave Kemeza) talked about their switch from PBS to SLURM. He reported that the switch more or less had been painless. After the NASA talk we talked with a SLURM and a SchedMd guy, asking them how SLURM performs under heavy load (1M jobs), whether SLURM supports automatic resubmission, if SLURM is able to correctly report CPU time for jobs which are forking and whether it is possible to simulate heavy loads with SLURM. He suggested that you can run many instances of slurm on the same machine to simulate a big cluster.

- nVidia At the nVidia booth it was possible to learn CUDA programming in 45 minutes, otherwise nothing interesting.

- Moab We also stopped at the Moab booth to hear about their LRMS and scheduler. They said that it could easily handle 100k jobs but above that level it started to have problems. Instead for larger amounts of jobs they suggested Nitro which is a sub-scheduler, which really does no scheduling, and runs on a dedicated node.

- ComputeCanada (globus) ComputeCanada is the collaboration of HPC centres in Canada, and they were looking into sharing data using the globus software.

- A*START Singapore (Galaxy) It is NOT Grid. It is NOT Cloud. It is a Galaxy of SuperDuperPuders. A poster showed that they used a long distance infiniband from Singapore to Seattle in order to connect HPC centres. Also from the poster it seemed they connected HPC centres in Singapore by infiband. They claimed to have created a unified distributed HPC.

- Faunhofer (BeeGFS)

-Energy Efficient HPC Working Group http://eehpcwg.lbl.gov/-

We talked briefly with a couple of representatives of the working group. One from the US and one from Germany. They would be happy to have more participants in their working group(free). Some of their problems was of a more practical nature. The US guy (NCSA?) had the problem that they had to call the local power utility when they started running jobs. The cluster would go from 6 MW idling to 13 MW under load, so they could not just start jobs on all machines, but would have to slowly inject jobs building towards maximum load. On their site they have a white paper on a standard methodology for measuring power consumption, which could be useful for Exercise.

next year

There was no booth selection meeting this year. It is possible to request favorite booth locations on the contract, but there will be no guarantee.

There was no reminder sent that power must be ordered seperately (a new procedure). You must remember to do this for next year.

You must be at the convention center on Sunday to check on the status of the booth. Power and carpeting must be fixed before the main carpeting is installed Monday from 8:00-12:00.

We need some kind of folder that explains what ARC is and what it can be used for.

Last years gadgets were good to get rid of business cards. How many contacts did we gain from this? Hard to tell since we didn't have email on card.

Pictures

Actions

Read through the exhibitor info

DEADLINES

Action Status/Deadline Comment AssignedTo
Find People Done - Farid
Passports, Visa etc Done - Martin, Christian
Submit Company Profile Done Profile from last year used
Pay for the booth Done USD 100 paid in April 2014 Farid
Order furniture Done Furniture has been ordered Oxana
Order electricity Not done One 110V outlet is included in the furniture order, see above. Apparently one have to fill a separate form to get electricity - not the same form as the freeman one. See notes above.
Order Network Done Not Ordered -
Tickets Done - Christian
Hotel Done Hotel Helene all
Design Poster NA Existing posters used
Print Posters NA Existing posters used
Print Address cards NA Address cards not printed
Bring Posters and address card Done - Christian, Martin
Prepare slide show NA Not going to show slides
Prepare gadgets NA No gadgets
Allocate notebooks Done - Christian, Martin
Borrow a beamer NA -
Borrow tripod for posters NA Optional
Poster fixation on curtains Done -
Prepare scotch- and brown tape Done -
Register booth personnel Done costs 100 greenbucks
Take socket adapter, power cord distributor Done -
Hire a car Done - Christian
Box cutter, straight threshold profile, scissors Done -
Cell phone + charger Done - Christian
Submit booth space contract for SC15 Not done - Farid
Create SC15 wiki page done - Christian