Special Report on Super Computing 2017

Super Computing 2017 is not a conference dedicated to storage but storage is pretty well represented and without storage this industry can’t exist.

It’s true for HPC and for all other sectors finally, as storage is our memory and you can’t rebuild or recompute stuff everyday and some of these data can’t be generated as they come from acquisitions methods… thus we saw storage popping up in many different places.

The 2017 edition took place at the Convention Center in Denver, CO, with around 12,000 attendees from 71 countries with 2,500 coming from outside USA and 334 exhibitors globally as vendors and universities/research centers. 122 companies are international ones. With 78 workshops and tutorials, 127 posters, 77 birds-of-a-feather this event is also an education moment for the community that connects well together.

In term of booth, the size is a good indicator of the HPC involvement for the company. Nvidia has the biggest one followed by Intel, Dell EMC, HPE, ExaScaler, Mellanox, DDN, Seagate, Lenovo, Cray, Penguin Computing, Google and Huawei.

The conference had a dense program and the top topics were GPU, of course, burst buffers and fast I/O, file systems/storage, NVMe and composable infrastructure. HPC developments are more and more present in the more ‘classic’ IT solutions and our day to day IT life.

As usual, the Top500 announcement was a key moment, this year it was the 50^th, and China overtakes USA with 202 HPC systems vs. 144. China takes the lead with the first two places for biggest HPCs with Sunway TaihuLight at NSCC in Wuxi and Tianhw-2 at NSCC in Guangzhou, Switzerland is 3^rd with Pix Daint, 4^th is Japan with Gyoukou and finally 5^th is USA with Titan.

It was also the first conference where the IO-500 list managed by the Virtual Institute for I/O was unveiled. Among the top 9 systems, 3 uses Lustre, 2 Spectrum Scale and BeeGFS, 1 use DataWarp from Cray and the top one relies on DDN IME.

As a leader in HPC Storage, DDN has demonstrated and announced interesting product iterations: new IME entry level (IME140), a monitoring tool (DDN Insight) and a new declustered RAID aka DCR for SFA14KX that provides better throughput and faster rebuild times.

Cray, with its recent acquisition of Seagate ClusterStor line, introduced the L300N hybrid product dedicated to mixed workloads powered by NXD, a new flash I/O accelerator software for Lustre.

HPE has made a great announcement with DDN illustrating the credibility they wish to reach and holes in the HPC product line.

We tried to obtain information about Weka IO on the HPE booth and it was a terrible moment, not for us but for several HPE people unable to give us any comments, many of them don’t know the company name. In fact, there were no posters, slides, data sheets… about Weka IO software on HPE booth, just the talk by Liran Zvibel, CTO and cofounder of Weka, on the booth theatre. We have to recognize that HPE didn’t announce anything with Weka IO as the PR came from Weka IO but that PR contains a quote from Bill Mannel, vice president and general manager, hopefully. We also understood that HPE had to protect their DDN partnership.

We announced recently the departure of Michael Raam, CEO of Weka IO, and finally the company has reacted fast with Liran Zvibel taking the position.

Last funny point: HPE renamed DMF into Data Management Framework, who meant for decades Data Management Facility coming from Cray then Silicon Graphics and then SGI.

IBM’s booth demonstrated interesting products but we were surprised not to see Cloud Object Storage (COB), coming from Cleversafe acquisition, when IBM promotes its own data protection approach with Spectrum Protect. On the other hand and especially in the file storage area, the company pushed Spectrum Scale (aka GPFS), Elastic Storage Server and COB, which is a second surprise.

Quantum did a good promotion for StorNext, of course, but also of Rook, a multi-protocol SDS – file, block and object – product for massive volume of data, based on Ceph.

Panasas entered in a new era after several years of re-design and re-architecture of the product. The company has announced the 7.0 major PanFS release and a disaggregated model.

Flash and NVMe storage was well represented even if pure player presence was limited. Among others we saw HGST, Pure Storage, Seagate, SmartIOPS and Vexata and the rest was more global vendors such DDN, HPE, IBM, Inspur, NetApp, QCT or Super Micro.

Among the new file storage vendors  – vendors with a innovative distributed file system – we noticed the presence of Avere Systems, Elastifile, Panasas and Qumulo but Rozo Systems and Weka IO didn’t have a booth. And of course, we saw the classic presence of Lustre- and Spectrum Scale-based file storage solutions.

On the object storage side, it was limited as pure players were pretty much absent like Cloudian, except Caringo. We noticed the presence of the technology at some booths as slides or pod such HGST with ActiveScale and partnership with Globus or configurations with Pixit Media Ngenea HSM and ArcaStream, Quantum with Rook, Spectra Logic with BackPearl, DDN with WOS, Lenovo with Ceph, NetApp with StorageGRID, Pure Storage with FlashBlade or IBM with COS but many other vendors didn’t mention at all their object storage offering like Fujitsu who didn’t show CD10000 S2. It confirms two things again: object storage is a capacity tier and file access is king.

Also, Atempo has exhibited as an independent company following its cut from ASG a few months ago.

The presence of Space Monkey was a surprise as the company was acquired by Vivint and is fully outside HPC. Space Monkey is a dispersed P2P storage.

Los Alamos National Laboratory announced also some efforts around BitScope, a 5-rack cluster of Raspberry Pi boards, with 750 CPUs or 3,000 cores to facilitate HPC R&D based on highly parallelized inexpensive approach.

Next year the event will take place in Dallas, TX, from 12 to 15 of November, 2018.