Pure Storage FlashArray systems are all solid-state systems, providing both FC and iSCSI block storage access. All data is both deduplicated and compressed prior to being written. Pure Storage claims there is no performance penalty for deduplicating and compressing data in-line, which enables an effective capacity ranging between 61 TB and 3.3 PB for their //X products and up to 5.2 PB for their //C product.
With FlashArray //X, Pure Storage has updated their second generation //m systems that continue to utilize purpose-built controller hardware and a custom NVRAM module. The system has dual controllers with either SAS attached SSDs or DirectFlash modules internal to the controllers and add-on device enclosures. In 2019, Pure Storage introduced the FlashArray//C system as a high capacity flash array utilizing QLC flash.
Evaluator Group Comment: Pure Storage marketing calls the controllers “active/active” but they are really active/passive in that both controllers cannot be actively transferring data between devices and hosts. The “expanded” definition of active/active used is because the standby controller keeps the host bus adapters active and allows selections to pass to the other controllers.
The FlashArray //X is an update of the controller hardware but the first that utilizes NVMe protocol and PCIe interfaces for custom designed flash modules called DirectFlash. DirectFlash can replace the use of standard SSDs in FlashArray //X, FlashArray //C, and previous //m systems are upgradable according to the guarantee made in the fall of 2016. Pure Storage claims a 4x performance improvement with the use of NVMe DirectFlash modules vs SSDs. The //X70 and //X90 can utilize a further performance increase by using Pure Storage’s DirectMemory cache. DirectMemory offers Optane SCM modules that can be used for read caching.
Use of NVRAM accelerates write performance by allowing the system to acknowledge a write and then coalesce data before storing it across solid-state drive modules.
The embedded “Purity OS” now at version 5.1 (FA5) has many advanced features including encryption, snapshots, and remote replication along with a vCenter web plugin and VMware VAAI support. Version 4.8 added a fairness algorithm called Always On QoS to eliminate the noisy neighbor problem and NPIV support for upgrades to redirect I/Os to ports. Version 5 added VVOL support, improved compression, active-active stretched cluster support, the ability to run VMs and containers on the storage platform, and portable snapshots including CloudSnap.
Dual controllers connected with Intel PCIe bridging provides HA, along with incremental performance due to round-robin volume access to host volumes over multiple paths. Capacity may be scaled by adding additional storage trays, with up to 4 trays. Pure Storage increases the effective capacity with deduplication and compression and achieves an average data reduction ratio in excess of 5:1 according to field data.
Incoming data is written to NVRAM, and mirrored to the partner controller in HA configurations. Next, a write acknowledgement is sent to the client. A unique hash is created from the data and then stored at the address of its computed hash. This two-tiered write caching helps to alleviate host visible write delays encountered from SSD devices when performing garbage collection.
By storing data at an address dependent upon its content data is deduplicated, with duplicate data stored at the same address. Finally, data is de-staged to flash in batches of several megabytes at a time. Each flash module is enabled to either read or write data, but not both simultaneously, since that can have a performance impact.
Three copies of metadata are stored with dual parity protection, providing high data integrity. Additionally, read requests are possible from partial RAID stripe reads, by reconstructing data inline if sufficient data has been read.
The Pure Storage //X systems are designed as a general purpose, high performance storage system competing with other mid and high range storage systems. Target applications are those requiring high performance without a large capacity requirement.
The //C arrays are focused on high capacity storage tier 2 storage applications. Use cases for the FlashArray//C include policy-based tiering, disaster recovery, and snapshot consolidation.
The embedded Purity OS adds advanced features, Fibre Channel and iSCSI host access, NVME-oF support via RoCE, and active / passive HA controller nodes. The systems are highly available, providing all hot-swap components and symmetric LUN access. This enables non-disruptive access during component failures and upgrades, including for host access paths or storage controllers.
VMware support enables users to perform storage allocation along with performance and capacity management from within the VMware console. Additionally, support for VMware VAAI provides enhanced performance with VMware based workloads, including thin reclamation, block zeroing and virtual machine cloning operations.
- Performance – Pure Storage provides their own performance numbers with the //m70 system at 300,000 32K IOPs and 9 GB/s bandwidth. With the introduction of model //X and the NVMe protocol DirectFlash modules, Pure Storage claims a 4x performance increase vs pervious systems using SSDs. Read latency can be increased with the use of DirectFlash SCM.
- Availability – With two controllers, the FlashArray has the high availability characteristic including the failover and failback required.
- Replication for BC/DR – Both asynchronous and synchronous remote replication is available.
- Data reduction – inline compression and deduplication is always on with the system.
- The Pure Storage FlashArray //X is targeted at applications that use block I/O with high performance requirements. The FlashArray //C targets high capacity needs and tier 2 applications.
- The features of snapshot and remote replication give customers high-end system capabilities. The performance and features allow applications to be deployed in a high availability environment.
- System environments
- The system is targeted for SAN storage with NVMe over Fabric – RoCE, Fibre Channel, and iSCSI environments.
- Deployment and Administration
- The system is intended for installation and management by the IT generalist and includes wizards and quick start configurations to allow for simple deployments.
- The system reports to Pure Storage web software for continuous analysis of log data and the ability to use a SaaS application for monitoring and reporting.
Evaluator Group EvaluScale™: Pure Storage FlashArray – SAN Storage
Evaluator Group product review methodology “EvaluScale” assesses each product within a specific technology area. The definitions of the criteria and explanations of how products are reviewed can be found in the Evaluator Series Evaluation Guide. Download the product brief now to view the EvaluScale for this product.
Evaluator Group Opinion: Differentiating elements for Pure Storage FlashArray
Pure Storage FlashArray was designed as an all-flash system using commodity servers and SSDs. As one of the early flash systems to market, Pure Storage has done well with systems that provide good performance with modest capacity, increased by software-implemented compression and deduplication. Pure Storage has also continued to develop the system with improved features and adding support for NVMe and PCIe attached devices, with a 4x performance improvement according to Pure Storage information.
The system is a traditional dual controller design with active/passive for data transfer between devices and hosts. Pure Storage claims active/active because the HBAs on the standby controller remain active for selection but do not transfer data. It is unfortunate that the claim of active/active is made because it gives competitors the opportunity to point out how misleading Pure Storage is and detracts from the real story of how valuable the products have been. Pure Storage has been very competitive with price and has performance to meet most customer demands.
The Pure Storage FlashArray is not a scale-out system, which means for increasing performance in parallel with capacity, a new controller pair (another model) must be installed. Other competitive systems have implemented scale-out functionality for a more measured growth. Evaluator Group believes that eventually Pure Storage will continue advancements to counter the competitors that have scale-out systems. Other developments such as stretched clusters and further integration of other systems software have made the Pure Storage system more widely applicable as a solution for primary storage. The transition to NVMe and PCIe attached devices will give Pure Storage a competitive lead in performance.