Pure Storage Logo on a FlashArray //C
Pure Storage Logo on a FlashArray //C - Source: Max Mortillaro

On 18th September 2019, Pure Storage announced several new products at their yearly Accelerate conference, held this year in Austin, Texas.

Pure Storage Cloud Block Store

Making its debut here in Austin, Cloud Block Store (CBS) is a cloud-native block storage solution running on AWS. Most of what you may know in terms of cloud-based block storage are ported solutions running as a virtual appliance.

CBS takes a radically different approach: it uses cloud constructs to implement a storage architecture that benefits from cloud-native technologies. Pure have looked at the components of a FlashArray and implemented them in the cloud using various types of EC2 instances combined with S3 storage.

Architecture

The solution comprises of:

  • 2x EC2 instances acting as CBS Controllers (similar to a physical storage controller)
  • 7x virtual drives, each composed of:
    • An EC2 instance
    • IO1 EBS (AWS Elastic Block Storage) used as NVRAM and as a write buffer
    • A non-persistent Instance Store used as a Read Mirror
    • All of the drives are grouped into a virtual shelf (logical construct)
  • S3 is used as the durable & persistent data repository, with high bandwidth restore capabilities
Pure Storage Cloud Block Store Architecture
Pure Storage Cloud Block Store Architecture – Source: Pure Storage Marketing Slide Deck

From an availability perspective, CBS is highly available, located within a single Availability Zone, and uses APIs and functionalities that are consistent with a Pure Storage FlashArray. EC2 auto-scaling instances are also used; if a virtual drive is lost, another one is provisioned and rebuild happens.

Use Cases

No doubt that the feature set will help support use cases promoted by Pure Storage. The major ones include:

  • Cloud-based Disaster Recovery
  • Lift & Shift migration: moving to the cloud existing on-premises workloads leveraging block storage, and continuing to provide them the ability to consume block-based storage – Lift & Shift is an approach often taken when decommissioning aging / outdated infrastructure; the same band-wagon VMware and Oracle are jumping on.
  • Dev/Test use cases, where cloud elastic consumption capabilities make more sense than building an on-premises environment

CBS also supports Pure’s ActiveCluster feature, so it’s perfectly OK to build two CBS in different Availability Zones and enable ActiveCluster. In that case, Pure1 Cloud can be used as a witness for the solution.

“Cloud block store uses cloud constructs to implement a block storage architecture that benefits from cloud-native technologies”

Pure Storage FlashArray //C

The other major announcement is the immediate availability of the FlashArray //C, the latest addition to Pure Storage’s product line.

Architecture

The //C uses the same constructs as the //X. It has a similar chassis, uses the same DirectFlash modules and is operated by same Purity operating system. It is however intended to be used as a secondary storage platform.

This is achieved using a combination of QLC 3D NAND and Intel Optane Storage Class Memory (3D XPoint). The FlashArray //C is optimized for capacity and will be completely compatible in terms of feature set with the existing other products.

Although the //C uses the same technology as the //X, the way reads & writes are handled differ. This is due to the higher latency of QLC memory and to the lower endurance of QLC (<1000 P/E cycles).

Pure Storage FlashArray //C capabilities
Pure Storage FlashArray //C capabilities – Source: Pure Storage Marketing Slide Deck

Use Cases

This makes the //C an ideal platform for secondary storage use cases such as:

  • Disaster recovery for Tier-2 applications (application with lower performance requirements), i.e. running Tier-2 from a less expensive FlashArray //C at a remote location if the primary location with a FlashArray //X is lost
  • Snapshot replication between a Tier-1 FlashArray //X to a Tier-2 FlashArray //C
  • Policy-based data movements to the //C (as a cheaper, capacity based storage platform)
  • Backup target, to store backups from apps running on the //X into the //C (with appropriate data management software, policies can also help move older backups to a cloud-based archival tier (such as S3 or Glacier)
  • Dev & Test use cases, where the data is stored on the //C to keep Tier-1 storage capabilities of a FlashArray //X available for production workloads

Interestingly, the //C was designed to address a sustained demand from existing Pure Storage customers who were lacking a consistent end-to-end experience across tiers.

The //C is generally available now, and will initially ship with TLC memory, with expected shipment of QLC-based models in early 2020. From a pricing perspective, TLC models will be priced as if the capacity was provided with QLC flash.

“pure storage took the radical approach to replace hard drives with qlc flash, a step the industry may soon be following”

Pure Storage Direct Memory

Pure also introduced DirectMemory, a high-speed read caching technology developed by Pure whose aim is to reduce read latency for high-locality, performance critical applications (high-throughput, latency sensitive workloads). By high-locality, Pure intends workloads that often reuse a dataset fitting in the cache.

DirectMemory relies on Intel Optane (3D XPoint) Storage Class Memory NVMe drives. The latency of those drives is in the 10 μs range, making them ideal for caching. These modules are inserted into the chassis and exist either in 3TB or 6TB modules.

A Pure Storage DirectMemory Module sporting Intel Optane flash memory
A Pure Storage DirectMemory Module sporting Intel Optane flash memory – Source: Pure Storage Marketing Slide Deck

From a benefit standpoint, 80% of arrays can achieve at least 20% lower latency, and 40% of arrays can achieve from 30% to 50% lower latency.

There is one use case for DirectMemory: improving the speed of read-intensive applications that frequently need to access the same dataset. DirectMemory thus works ideally with in-memory platforms & databases (for example SAP HANA, Couchbase & VoltDB) supporting business-critical workloads.