All-flash arrays, and in particular those from startups, have been the stars of the flash revolution. With architectures designed from the ground up for flash, they have pioneered rapid-access, solid-state storage for the era of the virtualised datacentre.
But three or four years on from the onset of flash, we still find most customers do not have all-flash arrays installed, and instead have opted to build hybrid flash arrays from existing systems.
There are some key reasons for this.
All-flash arrays often offer blistering performance – up to and exceeding one million input/output operations per second (IOPS) – but this comes with a hefty price tag that means they are out of the range of all but the most performance-hungry, deep-pocketed organisations. A mission-critical, money-no-object application is needed to justify the spend.
In addition, most all-flash arrays are likely to end up as standalone subsystems, unable to work seamlessly with existing storage hardware. They are also unlikely to have advanced storage features, such as replication and snapshots that can work with other suppliers’ arrays that are in place.
And so, more often than not, customers have chosen to add flash to existing systems. This has the benefit of building on their investment in incumbent arrays, while being able to take advantage of the benefits of flash.
More on flash storage
- All-flash arrays startups survey 2014
- All-flash array roundup: A two-track approach from big storage
- All-flash arrays: Will time run out for mainstream acceptance?
- Flash storage: What it is and where to deploy it
- What's wrong with flash storage: And what will come after?
- MLC vs SLC: Which flash SSD is right for you?
Sure, this approach doesn’t usually allow for hyper-fast performance but it can direct low-latency storage to where it is needed for key applications. Often this will be carried out by the use of automated storage tiering that pushes the hottest data to a flash tier that comprises a small percentage of overall capacity.
Alternatively, volumes can be pinned to flash for performance-sensitive apps that always need flash’s low latency, whether data is hot or not.
So, for the time being, flash in a hybrid array from an incumbent supplier is the way most people choose to implement solid-state storage.
Here we cast a glance over the storage area network (SAN) products available from the biggest six storage suppliers and see what they offer.
Most rest content for now, with the ability to offer flash drives as part – or sometimes all – of total capacity, without any upgrade to hardware that can speed processing or backplane traffic for solid-state storage.
But others have developed or inherited controller hardware more suited to the needs of flash, such as EMC, Hitachi Data Systems (HDS) and HP.
EMC’s enterprise SAN VMAX family and mid-range to entry-level enterprise VNX arrays can all be configured with up to 100% flash drives.
VNX arrays benefitted from a controller software upgrade in late 2013 that allows them to take advantage of Intel multi-core Xeon 5600 central processing units (CPUs), as well as optimisation of array hardware for flash storage, using the latest Gen 3 peripheral component interconnect express (PCIe) cards in the controller.
Flare, the previous VNX operating system (OS), wasn’t built for multi-core CPUs and suffered from processing bottlenecks when using flash drives. It was rebranded as MCx and rewritten to spread workload across up to 32 cores in the Xeon processors.
While all the VNX arrays are optimised for flash with the MCx upgrade and could be entirely populated with flash drives, EMC also sells a VNX array marketed as all-flash called VNX-F, which scales to 400TB of capacity. EMC claims the VNX7600-F can handle 500,000 8K IOPS.
NetApp recently moved to phase out some of its fabric-attached storage (FAS) filer lines and replace them with products that can be hybrid flash or all-flash.
In particular, it announced it would phase out the FAS2000, FAS3000 and FAS6000 families and replace them with the FAS8000 and FAS2500 series.
All of these come with Flash Pool – a few TB of Data ONTAP-managed auto-tiered cache – while the FAS2500s (launched in 2014) come with the option for more than 50% of drives to be flash.
The FAS8080EX and FAS2520 can be all-flash. But, the other arrays in the FAS2500 range are limited in the proportion of flash they can accommodate, with a maximum of 96 out of 144 drives able to be flash.
The older FAS8000s can only take a few TB of flash capacity on arrays that scale to the single-figure PB.
NetApp’s E-Series arrays have one all-flash array – the EF550 – but lack any reference to flash drives in their spec sheets.
The Dell Compellent SC4020 combines Fibre Channel and internet-small-computer-system-interface (iSCSI) access, and can be hybrid or all-flash, with multi-level cell (MLC) or single-level cell (SLC) drives, as well as serial-attached SCSI (SAS) hard-disk drives (HDDs) up to 15,000 revolutions per minute (rpm). The base unit can hold 24 drives and the system can scale to 120 drives with added disk enclosures.
Capacity can reach around 500TB, while performance tops out at around 120,000 IOPS and latency of “less than a millisecond”.
Dell also makes the Dell Compellent SC220 Flash-Optimised Solution, which is a disk enclosure add-on to its SC8000 controller that can accommodate 400GB or 1.6TB MLC or SLC flash drives, as well as an SAS spinning disk.
In its iSCSI-only EqualLogic range, Dell has the PS6210XS, which is a hybrid flash model that can house up to 26TB, comprising seven 2.5in 800GB flash drives and 17 10,000 rpm SAS drives.
The PS6210 family was introduced in 2013 as successor to the EqualLogic PS6110 series and features upgraded CPUs, increased memory, a doubling of the 10Gbps ports and a new version of the controller OS.
HDS’s foray into flash so far comprises the Hitachi Accelerated Flash Storage (HAFS) module for its enterprise SAN virtual storage platform (VSP) and Hitachi Unified Storage (HUS) arrays. HAFS is a controller with a custom application-specific integrated circuit (ASIC) and firmware designed for MLC flash.
HAFS comes in 2U modules with 2.5in 400GB or 800GB MLC flash drives with a maximum capacity of 19.2TB, or a dense array with HDS’s proprietary Flash Module Drives (1.6TB or 3.2TB) that provide up to 38TB of MLC capacity.
Up to four flash enclosures can be housed in a VSP array. VSP can also treat HAFS as a distinct tier of storage using its Hitachi Dynamic Tiering.
HP has hybrid flash options across nearly all its storage array product lines – 3Par, StoreVirtual, MSA and XP.
HP’s enterprise-class 3Par products are built for high-capacity and cloud operations, with features such as multi-tenancy and data tiering. The 3Par line is not a ground-up flash design, but the 3Par OS and controller ASIC have flash-friendly features that help reduce wear and provide the fine levels of granularity suited to flash cell block sizes.
3Par arrays include one all-flash array – the StoreServ 7450 – and two that can be configured as hybrid flash, the 7000 and 10000.
The StorServ’s 10000’s two variants scale to four and eight controller nodes. Capacities are 1.6PB and 3.2PB in 24 or 48 disk enclosures, with MLC and enterprise MLC (eMLC) drives mixed with Fibre Channel or SAS HDDs, as well as 8Gbps or 10Gbps Ethernet connectivity to hosts.
The StorServ 7200 and 7400 slot into the range below the 7450 all-flash array. They have either two (7200) or two to four (7400) controller nodes and scale to 240 or 480 2.5in drives and have Fibre Channel and Ethernet connectivity.
HP’s StoreVirtual family – built on internet protocol (IP) acquired from Lefthand Networks in 2008 – are iSCSI block-access products. One product in the range, the 4335, can accommodate a proportion of flash solid-state drives alongside SAS HDDs, and scales from 7.5TB to 240TB.
HP’s entry-level Fibre Channel and iSCSI MSA family includes one hybrid array, the 2040. It can house up to 199 2.5in flash drives or SAS HDDs – or 96 in 3.5in format. Maximum capacity supported is 384TB, which is accommodated in added disk enclosures.
HP’s XP high-end storage arrays – repackaged HDS hardware with HP software – scale into the hundreds of PB and have mainframe fibre connection (Ficon) connectivity alongside Fibre Channel and iSCSI. XP arrays allow for a proportion of total capacity to be flash SSD, alongside an SAS spinning disk.
IBM’s DS8000 series enterprise arrays are for high-end use cases and compete with EMC’s VMAX arrays. They offer all-flash and hybrid-flash options and scale to PB levels with Fibre Channel and Ficon access. Flash drives can be added alongside a spinning disk or in a separate flash enclosure, the DS8870.
XIV is an enterprise-class set of arrays acquired in 2008 by IBM from an Israeli startup. It was designed to support cloud environments and more recently has been targeted at analytics workloads.
XIV has a parallel hardware architecture with multiple controller nodes – while data is distributed for availability and data protection with rapid drive, rebuild speeds are claimed. XIV is connected to iSCSI and Fibre Channel, and offers a small portion of flash caching per XIV module.
IBM’s Storwize V7000 arrays are attached to Fibre Channel and iSCSI, with a unified model that also provides NAS functionality. They scale up to 1,056 drives in a clustered setup and can accommodate flash drives alongside SAS or a Nearline-SAS spinning disk.