nav search
Data Center Software Security Transformation DevOps Business Personal Tech Science Emergent Tech Bootnotes BOFH

Tegile gets IntelliFlash into a cluster duster

Cloud Platform for private cloud

By Chris Mellor, 29 Aug 2016

Tegile is launching its all-flash, clustered, IntelliFlash CP (Cloud Platform) at VMworld 2016.

This is a rack-scale system for large enterprises looking to set up a private cloud and, Tegile says, involves multi-controller clusters and NVMe. Individual systems are clustered together with a single global namespace. They will have different grades of media with intelligent data placement.

IntelliFlash CP provides scale-out growth and supports multiple protocols.

The original IntelliFlash HD (high density) was announced a year ago and based on SanDisk InfiniFlash JBOF (just a bunch of flash drives) hardware, which offered 512TB of flash capacity in a 3U enclosure. This was built from 32,768 128Gb 15nm MLC NAND SanDisk die spread across 64 x 8TB proprietary flash modules.

Tegile said there could be eight 3U IntelliFlash units in a rack, with 18U needed for controllers and a rack switch, providing 10PB of effective capacity after data reduction and a $0.50/GB effective capacity price.

IntelliFlash HD supported two grades of flash media: a performance tier and a capacity tier. Its IntelliFlash OS supports snapshots, clones, thin provisioning, inline deduplication and compression. It has VMware integration via SRM, VAAI, VASA, and VVOLs and it supports both block (FC, iSCSI) and file (NFS, CIFS/SMB3) access.

A year ago, Tegile said IntelliFlash would support 1PB in 3U in 12 months time. That hasn't happened yet. There would be a move to NVMe drives inside the array followed by a move to NVMe over Fabrics to link to accessing servers, making this a DSSD kind of system. NVMe drive support is happening but not yet in the InfiniFlash part of the system.

IntelliFlash CP speeds and feeds

The company says; "IntelliFlash CP lays the foundation for an integrated system that will use NVMe to massively reduce network latency to meet the enterprise need for real-time access to data. By leveraging NVMe to connect SSDs to the controller plane, IntelliFlash CP will be able to provide memory-to-memory transfers nearly as fast as accessing local file storage at sub-millisecond latencies. This is unprecedented latency in the storage industry."

Narayan Venkat, Tegile's chief marketing officer, told us the SanDisk IF150 chassis, the capacity tier, has 512TB raw capacity and connects to the IntelliFlash controllers by 12Gbit/s SAS. There is approximately 350TB effective capacity per IF150 before dedupe. Effective capacities will vary based on dual parity or triple parity.

Each controller chassis has two controllers, providing high availability, and has 20 or 40 NVMe drives. These form the performance tier. Capacity per NVMe SSD starts with 800GB but will likely be 1.6TB by GA time. There is a maximum of eight nodes per cluster.

Venkat said the "cluster interconnect between nodes is 40GbE. Within each controller chassis the two controllers use NTB for inter-cluster communication. Nodes are easily added to the cluster using the cluster management UI or CLI. Storage pools, resource ownership and cluster expansion are automatically managed by the cluster management services in the system.

"Removing a node from a cluster is also simple. Storage pools are moved from one node to another. No data copy, just ownership, is moved. Since we connect IF150 JBOFs over SAS fabric, if both controllers in a node go down, we can failover storage pools/vols on to existing nodes in the cluster."

Venkat said: "We can string 2 x IF150s behind each controller pair. So max of 8U size (2U+3U+3U) per node. Flexible. Can start with one and add one more in a scale-up fashion. And/or choose to scale out to another node."

And cluster capacity? "Total max raw capacity in a cluster is 8 x (1PB + 64TB) assuming 1.6TB NVMe SSDs at GA. Effective capacity before dedupe and compression is approx 6.3PB (high performance + dense flash.)"

Faster network interconnect

Venkat thinks SAS-based JBOFs will eventually move to NVMe over fabric interconnect, but this will take time to become broadly available.

When IntelliFlash supports NVMe over fabrics then it should have the kind of NVMe over fabrics speed espoused by Apeiron, E8, Excelero, DSSD and Mangstor.

Tegile claims that, with IntelliFlash CP, "enterprises will be able to build massively scalable infrastructure with up to tens of petabytes of flash storage. Enterprises will experience the resulting performance gains and benefit from pricing that is better than public cloud offerings."

Stifel MD Aaron Rakers has discussed the InfiniFlash roadmap. SanDisk will follow the IF150 "with an IF750 system release later this year that integrates SanDisk's Ion Accelerator (Fusion-io) software. We believe investors could consider significant potential capacity increase as Toshiba / Western Digital move toward volume 64-Layer 256/512Gb 3D NAND production in 2017 (ie, same number of die would equate to 1PB at 256Gb/die and +2PBs of all-Flash capacity at 512Gb/die in the same 3U configuration."

Imagine that fuelling a future IntelliFlash upgrade.

Check out this URL for more info. IntelliFlash CP will be generally available in Q1 2017 with configurations priced from 50¢ per GB. ®

The Register - Independent news and views for the tech community. Part of Situation Publishing