IBM FlashSystem – Argues For Flash-Only Storage Systems

IBM FlashSystem Highlights

• Invests $1 billion in R&D
• Launches 12 Centers of Competence across the globe
• Launches the IBM FlashSystem 820 based on TMS’s flagship product and adding IBM software and management features
• Launches a version integrating its IBM SAN Volume Controller software for storage hypervising
• Argues that high-priced all-flash storage can help reduce overall system costs by cutting server and software spend
• This all-flash storage approach is just part of a wide storage strategy

IBM has consolidated its Flash-only appliance offerings, investing $1 billion in R&D following its acquisition of Texas Memory Systems and opening 12 Centers of Competence around the world to help customers plan to drop spinning disk from key workloads. You’ll want to learn more about its views on how such implementations will lower the total system costs, despite increasing spending on storage.

What Is IBM Announcing?

IBM’s launch in New York City last week focused on 3 main areas, which were:

• An investment of more than $1 billion in Flash R&D in systems and software; significantly more than the (undisclosed) acquisition cost of TMS
• The introduction of the IBM FlashSystem 820 – a 1U rack-mounted storage systems with 20TB usable RAID-protected capacity, featuring Variable Stripe RAID and 2D Flash RAID; other FlashSystem models available; an enhanced version bundled with IBM’s SAN Volume Controller adds the advantages of storage hypervising, including business continuity with Copy Services, Flash Copy, thin provisioning, real time compression and Easy Tier software
• The opening of 12 new Flash Centers of Competence around the world[1], which will help speed adoption with clients by providing Proof of Concept, workload benchmarking, tailored ‘IT optimisation assessments’ and TCO studies

It was joined by a panel of users including Kruger, Reuters, Sprint, ViON and my ex-colleague Dave Vellante from Wkibon. The FlashSystem 820 is essentially a rebadging of the TMS RamSan-820 with the inclusion of IBM-specific software and call-home services. It didn’t talk about pricing at the launch, but we believe the basic array costs around $150k.
IBM notes that disk drives are the poor relation among data centre devices, improving their performance on average only 1.2 times, compared with CPU performance (8-10 times), DRAM speed (7-9), network speed (100) and bus speed (20 times) over the past decade. It makes sense to be thinking about replacing the only computer devices which still have motors (and to reduce the number of fans needed to cool the data centre), however there are major cost and resilience issues for users considering moving to all-flash storage. No matter how many companies decide to go down this route, hard disks will be with us for many decades to come – witness the continued use of tape storage in our market many years after the possibility of replacing them with disk.

Spending More On Storage To Reduce Server and Software Costs

IBM recognises the additional cost of all-Flash storage, but argues that its use cuts overall system costs. In a comparison of a disk-based and all-flash based system it showed an increase in storage costs from $420k to $638k was offset against a reduction from $7.1m to $4.9m overall: higher storage utilisation leading to 17% fewer servers, 38% lower software spending due to the reduced core count, 35% lower operational support costs for server and storage administration and 74% lower environmental costs due to the reduced power consumption and footprint.
Its views were supported by customer references from Banco Azteca, InterSport and CloudAccess.net.

The Cost Of Securing Data In IBM’s All-Flash Approach

TMS has 20 years experience of supplying all-flash systems and IBM itself has many research projects in place to work on the use of flash in new-style SANs, including the use of its SAN Volume Controller in advanced storage hypervising. Its FlashSystem uses a number of features and functions to improve the reliability and durability of an all-flash approach. In particular:

• It uses enterprise Multi Level Cell (eMLC) chips as the basis of the arrays, which tolerates 10 times as many program/erase (rewrite)) cycles [2] than consumer MLC technology; we note that in its recent UK survey XIO found almost no companies prepared to use consumer grade SSD
• It has added a number of features to the arrays including variable RAID 5 striping at the die level across each set of 10 chips within flash modules, as well as a second layer of RAID across flash modules, and has used its research to address ”disturb errors” by adjusting voltage and timing

Most warranties for flash storage are adjusted downwards to take account of the limits of their programmable erase cycles in comparison with spinning disk. In IBM’s case the use of eMLC chips and its advanced technology allows it to claim a 5-year useable life cycle for most applications – no major difference with spinning disk, whose effective life span is restricted to the same by continuous improvements incapacity and price decline.

The IBM FlashSystem Mega System Potential

FlashSystem can be used to create a single-rack of storage with mind-boggling dimensions. In particular:

• 1PB of storage in a rack covering just one floor tile with 100 microsecond latency, 22 million IOPS feeding 210GB per second and using 12.6 kW of power (equivalent to a 200TB array of spinning disk)
• The equivalent of 315 racks of performance-optimised disks, or 630 racks of high capacity disk

It seems unlikely that any one will rush out to buy a system which costs $12 million (in our estimation) today, no matter how compelling the space and power savings.

Some Conclusions – All-Flash Can Improve IBM’s Storage Systems Share

IBM and other large storage system suppliers have declining revenues, despite advising users on the need to spend big to meet the challenges of Big Data. In the case of IBM FlashSystem it has (and continues) to invest in an advanced all-Flash approach, which it believes will be appropriate for both ‘systems of record’ and ‘systems of access’ workloads. We are particularly impressed with its SAN Volume Controller version of the array – it makes sense to have flash at the head of your heterogeneous storage pool.
IBM is using its advanced knowledge and 12 Centers of Competence to overcome the reluctance users have in adopting all-flash arrays and is countering the higher cost of the storage by pointing to the lower cost of systems in which it is deployed. We are particularly impressed with the international spread of its Centers of Competence, especially in Asia Pacific where it will have less competition from startup suppliers.
We think it can improve its storage systems revenues through its investments, although it is unlikely to sell these high-end products to many small companies.
However don’t be misled into thinking IBM as a vendor is moving to replace spinning disk: this all-flash approach sits alongside its XIV, Storwize, DS8000, SONAS, SVC, SmartCloud Storage and even tape approaches. There are many suppliers offering hybrid flash and disk and a few offering flash-only arrays. While it’s inevitable that flash will play a bigger role in data centre systems, the timing and nature of a vendor’s product strategy will make a big impact on the success of its overall storage business.