nice architecture. Here are my notes.. tl/dr... Optimized for cost/GB Not designed for the fastest performance Provides NFS or S3 access to storage. Will add more storage protocols in the future Designed using decoupled architecture similar to cloud architectures You can scale the backend storage independently from the frontend Frontend consists of many stateless controllers which add performance linearly Backend consists of JBOF (just bunch of flash) trays that communicate via NVMe over fabrics to controllers Starting point is 1 PB usable (with data reduction) Systems have 600 TB real usable capacity before data reduction Data reduction is competitive with HDD based architectures from a cost perspective Data reduction is done via algorithm that compares similar blocks to gain better data efficiency as opposed to comparing identical blocks Uses lowest cost QLC SSD to reduce cost of $/GB Architecture uses QLC today but is designed to use cheaper SSD/Flash technology as it becomes available Uses Intel Crosspoint (XPoint) persistent memory technology to ingest data then write it to QLC (or other cheaper media) at VERY wide stripes across hundreds of drives. Data is written to QLC via large stripes i.e. 500+10 instead of traditional RAID 6 6d+2p this accelerates the recovery time of drive failures as only a subset of drives need to rebuild Not clea4 on.. data encryption at rest, snapshots, clones, remote replication, microcode upgrades, non-disruptive upgrades, etc…
We joined Tech Field Day again this year and spoke on the topic of Speed + Performance + Cost. Check out the latest VAST technology th-cam.com/video/xFpm3TjGqrI/w-d-xo.html we'd love to hear your thoughts!
nice architecture. Here are my notes.. tl/dr...
Optimized for cost/GB
Not designed for the fastest performance
Provides NFS or S3 access to storage.
Will add more storage protocols in the future
Designed using decoupled architecture similar to cloud architectures
You can scale the backend storage independently from the frontend
Frontend consists of many stateless controllers which add performance linearly
Backend consists of JBOF (just bunch of flash) trays that communicate via NVMe over fabrics to controllers
Starting point is 1 PB usable (with data reduction)
Systems have 600 TB real usable capacity before data reduction
Data reduction is competitive with HDD based architectures from a cost perspective
Data reduction is done via algorithm that compares similar blocks to gain better data efficiency as opposed to comparing identical blocks
Uses lowest cost QLC SSD to reduce cost of $/GB
Architecture uses QLC today but is designed to use cheaper SSD/Flash technology as it becomes available
Uses Intel Crosspoint (XPoint) persistent memory technology to ingest data then write it to QLC (or other cheaper media) at VERY wide stripes across hundreds of drives.
Data is written to QLC via large stripes i.e. 500+10 instead of traditional RAID 6 6d+2p this accelerates the recovery time of drive failures as only a subset of drives need to rebuild
Not clea4 on.. data encryption at rest, snapshots, clones, remote replication, microcode upgrades, non-disruptive upgrades, etc…
Thank for saving me from suffering through this.
We joined Tech Field Day again this year and spoke on the topic of Speed + Performance + Cost. Check out the latest VAST technology th-cam.com/video/xFpm3TjGqrI/w-d-xo.html we'd love to hear your thoughts!
Anybody want to share there opinion or experience on VAST vs WekaFS ?
WekaIO CEO, L. Zvibel also gave a talk at this conference.
Haha, tough crowd.