Geoff Hough, Director Product Management, 3PAR
As late as the 1970s, most filling stations worked like this: upon arrival, you were directed to pull up to a specific platform according to the type of fuel you were purchasing, as pumping stations were designed to serve just a single type of fuel. Next, the service attendant would proceed to pump the fuel into your car. The attendant would then take your cash or credit card, note the fuel charge displayed on the pump, and process your payment using the cash box or a credit card imprinter. After receiving your change or signing your credit card slip, you were on your way.
Some may mourn those bygone days. After all, attendants would often wash your windshield for you! However, for the most part, this old filling station model was quite inefficient. Jams were likely to develop behind the pumps that were in demand at a given moment in the day. If the attendant and his card imprinter were otherwise occupied, you would wait until he could get to you. All of this resulted in greater fuel cost, more stations, and much more time spent at the pump. In addition, the security and accuracy of your transaction was a matter of human diligence and integrity.
What is the point? Consider that storage today is provisioned much like fuel was pumped 40 years ago. In today's datacentres, the consumers of storage are directed to an appropriate (and hopefully available) array where they must wait for their storage to be provisioned and then rely on the ongoing diligence and integrity of others to ensure the security of their data. In other words, system, application, and database administrators must rely heavily on storage administrators for routine storage provisioning and data security. Meanwhile, precious storage expertise is consumed with these routine activities, leaving less time for IT initiatives with higher value. Given our understanding of traditional datacentre processes and technologies, can a more efficient model even be imagined? Is the idea of "self-service” storage even conceivable?
Returning to our filling station analogy, one can see that the move to a more efficient "self-service" model was enabled principally by shifts in technology. Pumps were developed that could serve multiple types of fuel. Pumps were fitted with dual hoses, instead of one, enabling greater and more flexible use. Secure transaction processing capabilities — today's card readers — were embedded into the pumps themselves. The innovations and capabilities that we take for granted today have enabled a new and infinitely more efficient model that has changed the experience — and economics of fuelling — forever.
As for the idea of “self-service” storage, we should ask ourselves: What technical or procedural innovations are necessary to enable such a model in the datacentre? One requirement that immediately comes to mind is that storage provisioning would have to be made simple — so simple that storage consumers could properly provision storage for their applications without deep storage knowledge or configuration expertise. Conceivably, there may be a number of ways to achieve this, but the most obvious and efficient place to effect this simplicity is in the storage array itself (the “fuelling platform”). Some array vendors — mostly new guard players such as 3PAR®, Compellent, and EqualLogic — are far along in this respect.
Presuming a storage device easy enough for anyone to use, the next requirement would be to provide multiple and highly scalable levels of service for multiple users, much like the gas pump that evolved to support multiple fuel types and higher levels of use. In the context of data storage, this would mean abundantly and independently scalable levels of connectivity (Fibre Channel and iSCSI), performance (transactional and sequential), and capacity (Fibre Channel and ATA). Typically, these sorts of needs have been met only in high-end arrays. Of course, “high end” usually means high cost. Mid-range arrays offer lower cost, but scalability requires multiple devices. Also, mid-range arrays are optimised for single, not multiple, service levels — which returns us to the model whereby users must be directed to the most appropriate and available "fuelling platform." N-way clustered architectures provide an attractive alternative to the classic modular versus monolithic dilemma, but users may observe that quality of service can be limited. For example, EqualLogic only accommodates iSCSI drives and IBM® Nextra only accommodates ATA drives. The customer loses the ability to choose between premium and regular octane.
The last requirement for enabling “self-service” storage is one surrounding secure application and administrative segregation. One of the reasons that consumers and fuel station managers both became comfortable with the idea of "self-service" was that privacy and control concerns were adequately addressed for all parties. With the advent of specifically designed and integrated credit card readers, if desired, consumers could conduct their payment transactions safely and without interference while station managers kept control of the cash box and any non-routine (exception) processing. In storage, this becomes a very challenging requirement. Persuading different user groups to share the same devices and infrastructure has traditionally been an uphill battle for centralised IT departments and service providers. Politics and fears of compromised service levels become difficult barriers to overcome. Even more troublesome for a storage administrator is the idea of shared control. As the guardian of data availability and security for multiple applications and user groups, the administrator can only imagine the potential consequences of sharing device control among multiple parties. One inadvertently deleted or improperly exported volume, and it’s “game over”!
What is required in these circumstances is something similar to the functionality that a hypervisor like VMware® provides for server environments. So-called “virtual machines” are secure, so the administrator of one virtual machine cannot deliberately or mistakenly affect the integrity of another. Service levels are assured by the "master administrator," who sizes, groups, and sets rebalancing polices for virtual machines across various physical hosts.
If a virtual machine-like capability is the last piece of the puzzle for enabling “self-service” storage, what are the options today? Storage array partitioning, which provides administrative segregation, can be found in a handful of high-end arrays. However, while it may seem like a plausible solution, the drawbacks are many. First, as mentioned previously, there is the cost of high-end storage. There is also the complexity of administration, which makes it impractical for non-experts to use high-end arrays. A final impediment is that the partitioning schemes employed by high-end arrays are physically and coarsely defined. This means that a partition's service level can only be as good as the number, type, and configuration of physical devices that can be affordably assigned to it. Unfortunately, this provides little price/performance incentive for a given user group to share a device.
Another and more promising alternative for secure application and administrative segregation within a shared array is functionality that actually mirrors that of a true hypervisor. This technology is available today and allows a master administrator to create "virtual arrays" within a single physical device. Capacity limits, host permissions, service level parameters like RAID level and disk type, and authorised users are defined by a master administrator on a "virtual array”-by-“virtual array" basis. Security is provided by device control, which is tightly circumscribed for each user group. And just as a hypervisor virtualises physical components and allows them to be shared by multiple user groups, so does this "virtual array" technology. In fact, by aggregating "virtual array" workloads across all physical components (controllers, cache, drives, etc.), user groups are able to attain higher levels of service more affordably than they can in a dedicated environment — an attractive incentive to move to a shared infrastructure.
It is an odd but exciting prospect to contemplate the idea of "self-service” storage. At the time, “self-service” fuelling was met by some with trepidation and others with anticipation. No doubt, the same will be true for “self-service” storage. And yet the technology enablers have come into place for the first time. These include:
- Storage arrays and interfaces that are simple to operate by non-experts
- Multiple and highly scalable levels of service that can be delivered affordably in the fewest arrays possible
- "Virtual array" technology that provides secure application and administrative segregation with attractive cost and QoS advantages
3PAR is exhibiting at Storage Expo 2008 the UK’s definitive event for data storage, information and content management. Now in its 8th year, the show features a comprehensive FREE education programme and over 100 exhibitors at the National Hall, Olympia, London from 15 - 16 October 2008 www.storage-expo.com
Source: StoragePR
Seed Newsvine