With a flurry of startups and the beginnings – with EMC’s acquisition of XtremIO and IBM buying Texas Memory Systems – of a wave of big supplier consolidation, the all-flash array market is in a dynamic state.
Driven by the needs of virtualisation, flash storage has significant benefits over traditional hard drive spinning media by delivering very high input/output (I/O) performance, even with highly randomised workloads.
So, what is an all-flash array, which are the key suppliers and where is the flash array market headed?
What is an all-flash array?
The most obvious benefit of using flash in the array is in delivering very high I/O capability. Typically, devices on the market are capable of delivering from 500,000 up to one million input/output operations per second (IOPS) in a single footprint, with latency figures of one millisecond or less.
Many suppliers have put SSDs into existing hardware platforms as a way to deliver all-flash solutions. Although this option provides better performance than a hard drive-based array, it does not best exploit flash storage capabilities.
This is because traditional storage arrays were designed to optimise performance of the slowest component – the hard drive – through the use of caching and intelligent algorithms that understand HDD geometry and the physics of accessing spinning media.
By contrast, all-flash arrays have been designed from the ground up to work with flash media. This means providing sufficient back-end bandwidth capacity to cope with solid-state media, removing caching where it provides no benefit, and working with the wear characteristics of the flash drives.
Solid-state media has a finite lifetime, so array designs cater for wear levelling across all devices and for multi-layer redundancy in case of failure.
Read more about flash arrays
- All-flash array marketing heating up, but is consolidation coming?
- All-flash storage helps credit union speed core banking apps
- EMC pushes for multiple flash locations in datacentre hardware
- Tucci: FusionIO ‘darn right foolish’ for flash attitude
- Cisco UCS blade servers integrate Fusion-io PCIe flash cards
Finding the right flash array use case
Delivering high performance is not cheap and solid-state arrays are significantly more expensive than their HDD counterparts. Suppliers have attempted to mitigate this through the use of compression, thin provisioning and data deduplication that reduce the effective cost per terabyte of implementing all-flash solutions.
The high cost factor means all-flash solutions are more suited to environments where low latency and high throughput are required, but can be justified by the business savings generated.
Suppliers quote example use cases such as online gambling sites and financial trading where microsecond savings on transaction times translate directly into financial gain.
SSD arrays – especially the mid-tier offerings – are also well-suited to virtualised environments, including server and desktop infrastructures, where I/O requirements are consistently high and random in nature.
Not all flash arrays are the same
Suppliers can take one of three routes when designing flash array products:
- Using standard SSD drives;
- Using bespoke SSD drives;
- Using flash memory modules.
As we discuss each supplier offering, we will see how these implementations differ.
As well as varying the storage media, suppliers have chosen to either scale within a single array or scale using multiple nodes.
Flash array supplier round-up
Today’s supplier offerings can be classified as high-end, including those from Violin Memory, Texas Memory Systems, Whiptail and Kaminario, and more mainstream products, such as those from Pure Storage, SolidFire and Greenbytes.
Violin Memory has taken the route of designing its own flash memory modules known as VIMMs. Each VIMM is accessible through multiple controllers using its proprietary Violin Switched Memory (VXM) architecture, which provides a highly resilient infrastructure. The 6000 series arrays scale from 16TB to 32TB of raw capacity (12TB and 22TB respectively after Raid overhead) with 500,000 to one million sustained IOPS and write latency as low as 90 microseconds. The Violin 3140 array provides a greater capacity but lower IOPS option, scaling to 40TB and 120,000 IOPS in a single 3U device.
Texas Memory Systems (TMS) has been making solid-state storage arrays for around 20 years and was recently acquired by IBM. Earlier this year it released its latest offering, the RamSan-820. This 1U unit can provide up to 24TB of raw capacity using eMLC (enterprise Multi-Level Cell) technology. In terms of performance, the RamSan-820 can deliver up to 450,000 read and 400,000 write IOPS with a write latency of just 25 microseconds.
Whiptail has recently revamped its product offerings into the newly branded Accela and Invicta arrays. The Accela is a single 2U unit that scales up to 12TB (raw) and can deliver 250,000 write IOPS with 100-microsecond latency. The Invicta combines up to six Accela nodes to create a single 72TB capacity array capable of up to 550,000 write IOPS. Whiptail recently announced a partnership with Micron to use its solid-state drives across the entire Whiptail product range.
The K2 array from Kaminario provides scalability up to 100TB and claims greater than 1.2 million IOPS using MLC flash media. Scaling is achieved using multiple blade servers, each containing flash and traditional hard drives, with the HDDs providing redundancy in case of flash drive failure. Individual blades or “nodes” are connected using a 10GbE backplane. As with most all-flash solutions, external connectivity is via multiple fibre channel interfaces.
Pure Storage is into the second generation of its FlashArray product. The FA-300 series scales to 22TB of raw capacity and 100,000 IOPS at one millisecond latency. The platform consists of a 2U controller and is expanded using 2U 5.5TB storage shelves. Pure recently added support for 10GbE iSCSI, and uses space reduction techniques to provide a more attractive price point for its products.
SolidFire has taken the modular approach to scaling flash arrays with its SF3010 and SF6010 series devices. The SF3010 has a capacity of 3TB (raw); the SF6010 doubles this to 6TB, both using 10 SSDs in a 1U server form-factor. Multiple nodes can be linked together, with each node capable of delivering 50,000 IOPS.
The future for the flash array market
All the suppliers mentioned, except TMS, are startup companies looking to take advantage of high-performance computing environments.
As yet, we have seen few of the big six storage suppliers coming to market with anything other than traditional all-SSD arrays. The exceptions are IBM (acquiring TMS) and EMC, which acquired XtremIO, an Israeli startup that had no products or customers.
Although EMC has yet to bring any product to market, the XtremIO acquisition adds credibility to this new category.
Within the next 12-18 months, we can expect to see more rationalisation of this busy space, with further acquisitions from the big suppliers. All-flash arrays will continue to be a very fluid part of the storage industry.