Slide 1

Slide 1 text

5/6/20 The Sad State of Affairs in HPC Storage But There is Light at the End of the Tunnel InsideHPC Slide Cast – April 23, 2020 Robert Murphy Director of Product Marketing Panasas, Inc.

Slide 2

Slide 2 text

5/6/20 2 ©2020 Panasas, Inc. HPC Storage Landscape 2020 Major Changes “Lustre is No Longer Open Source” - Director of Research Computing at a major U.S. university •  A DDN and Cray captive file system now •  Diminished open source economics GPFS Curtailed at DDN •  No one else has stepped in •  Even IBM’s ability to step up is limited •  Dearth of GPFS expertise IBM Spectrum Scale License Changes •  Capacity license and premium editions have significantly increased costs •  Spotty support Panasas Adopts COTS •  Leadership performance •  Breakthrough economics •  Unmatched reliability, ease of management and great support Weka Retrenchment? •  Very expensive and resource intense •  Multiple key employee BeeGFS Adoption Cools •  Unreliable to the point of data loss •  L3 support latency from Germany •  Few data services limits use cases Isilon Issues •  Performance tips over at scale •  Spotty support within Dell •  Opening for Qumulo

Slide 3

Slide 3 text

5/6/20 3 ©2020 Panasas, Inc. Fastest Parallel File System at Any Price Point Built for Price/Performance, Engineered for Manageability 0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0 PanFS BeeGFS Lustre GPFS Panasas v. Competition GB/s/100 HDD PanFS on ActiveStor Ultra 1PB, 16GB/s, $150K Limited Time Special Offer Comparative Read Performance of PanFS v. Competitive Parallel File Systems in GB/s This means that Lustre, GPFS and BeeGFS need nearly 2X more drives (at 2X the cost) to achieve the same performance as Panasas.

Slide 4

Slide 4 text

5/6/20 4 ©2020 Panasas, Inc. What Makes ActiveStor Ultra So Fast? Fastest, Most Efficient Parallel File System Small Files: F/S on SSD Intent Log TXN Log Metadata: DB on NVMe iSCSI RPC Command Queuing Task Prioritization DataStore MetadataStore Object Orchestration Layer Large Files: F/S on HDD •  Small files are stored on low-latency flash SSD •  Large files are stored on low-cost, high-capacity, high-bandwidth HDD •  Metadata is stored on ultra low-latency NVMe SSD •  An NVDIMM intent-log stages both data & metadata operations (2,560X lower latency than NVMe) •  Unmodified data and metadata are cached in DRAM PanFS Adaptive Small Component Acceleration (ASCA)

Slide 5

Slide 5 text

5/6/20 5 ©2020 Panasas, Inc. Benchmarked Linearity Great Architecture Leads to Great Performance Advantages Linear scaling of aggregate performance Constant performance per node Consistent and predictable performance Linearity continues past 400 nodes 0 500 1000 1500 2000 2500 3000 3500 4000 4500 0 5000 10000 15000 20000 25000 30000 35000 3 4 5 6 7 8 Per ASU Throughput (Dashed Lines MB/s) Total Throughput (MB/s) ASU Count (4 OSDs / ASU) RAID6+ Scaling of ActiveStor Ultra Agg Write Agg Read Per ASU Write Per ASU Read

Slide 6

Slide 6 text

5/6/20 6 ©2020 Panasas, Inc. Scale-Out NAS – The Other HPC Storage Falls Over at Scale Advantages Convenience Reliability Support Disadvantages Performance Fall Off Expensive Dell EMC Lustre Parallel File System v. Isilon Scale-Out NAS* Isilon Scale-Out NAS Parallel File System Characteristic Scale-Out NAS Performance Fall-off Examples: Isilon, NetApp, Qumulo, Vast Data Processed Concurrently Per Day * “Reference Architectures Of Dell EMC Ready Bundle For HPC Life Sciences”

Slide 7

Slide 7 text

5/6/20 7 ©2020 Panasas, Inc. Purchase Criteria and Operational Challenges Disconnect Hyperion Survey Reveals Organizations’ Misalignment in HPC Storage Buying Decisions Downtime: Almost half of the surveyed sites experience storage system failures once a month or more frequently. Downtimes range from less than one day to more than a week. A single day of downtime costs from under $100,000 to more than $1 million.

Slide 8

Slide 8 text

5/6/20 8 ©2020 Panasas, Inc. Hyperion Survey Reveals Organizations’ Misalignment in HPC Storage Buying Decisions Purchase Criteria and Operational Challenges Disconnect

Slide 9

Slide 9 text

5/6/20 9 ©2020 Panasas, Inc. The Alternative: Panasas Mission-Critical HPC Storage Performance Without Risk Large Global Consumer Goods Product Customer • CAE COTS applications – Finite Element Analysis – Computational Structural Mechanics – Computational Chemistry – Statistical Analysis • Use Cases – Product, package, and process development – Physical testing avoidance 33 Overview 36 Reliability 8 YEARS ONLINE A reliable filesystem is cost-effective across the organization TCO goes well beyond the cost of hardware, support and administration cost “8 years of zero unplanned downtime” Panasas customer ASRC Federal

Slide 10

Slide 10 text

5/6/20 10 ©2020 Panasas, Inc. HPC Storage Comparison Only Panasas Ticks All the Boxes Panasas ActiveStor Ultra w/ PanFS 8 IBM Spectrum Scale Lustre BeeGFS Isilon NetApp High Performance Competitive Acquisition Cost Reliability Easy Management Timely, Quality Support Frustration Free

Slide 11

Slide 11 text

5/6/20 11 ©2020 Panasas, Inc. HPC Storage Summary •  Lots of changes going on, most are negatively impacting HPC customers •  Panasas is back with leadership performance and price/performance •  Chronic HPC storage downtime is costing organizations millions •  Organizations no longer need to risk their mission with unreliable storage •  Panasas is the mission-critical HPC storage alternative that ticks all the boxes

Slide 12

Slide 12 text

5/6/20 Confidential ©2019 Panasas, Inc. Thank you www.panasas.com