EMC VPLEX Announcement
EMC has announced VPLEX, a significant new technology that combines virtualization and distributed cache coherency software that allows multiple block-based storage arrays spread across a network to look like a single storage resource. The underlying technology was acquired by EMC as part of the Yotta Yotta technology purchase in 2008. EMC's federated storage vision is the full implementation of VPLEX.
What is VPLEX?
EMC has projected four stages in the evolution of VPLEX;
Stage 1: Local VPLEX is currently packaged as an appliance. Initially the major value proposition for VPLEX will be the same as IBM’s SVC, Hitachi’s USP V, HP’s SVSP/LSI’s SVP and other virtualization platforms including NetApp's V-Series and EMC's Invista. It will allow the set up of a virtualization front-end appliance that manages many back-end storage arrays. The front-end virtualizes the arrays by mapping the back-end, block-based storage onto its own file-based system. The major benefits are the simplification of storage allocation, the ability to utilize storage resources more efficiently across the arrays in the storage network and support for dynamic migration of data from one array to another. This allows a very significant reduction in the complexity and elapsed time for array commissioning and decommissioning, and some improvement in the utilization of installed storage assets.
Stage 2: Metro This differs importantly from other block-based virtualization appliance implementations in that longer metro distances up to 100km can be established between the VPLEX controllers, using direct fibre channel only connections. The storage resources are seen transparently across this network, allowing great flexibility in allocating and utilizing resources. All the VPLEX controllers are active in the network and have a copy of the distributed cache. Interesting topologies for creating synchronously replicated copies across metro network distances can be made with the potential for creating more flexible synchronous disaster recovery environments.
At EMC World, EMC demonstrated, using VMware vMotion, the transfer of applications dynamically between two metro sites across a VPLEX metro network. VPLEX, with its larger caches helps to speed up the transfer and is much faster than using vStorage motion. Operationally the transfer was transparent to end-users, with little impact on performance. The size of the databases in the demonstration was relatively small (26GB), but the speed and ease of use was impressive.
Stage 3: Asynchronous A future version of VPLEX will provide asynchronous support for moving data over long distances. Operationally, the network will look similar to the Metro topology. Asynchronous is restricted to two VPLEX nodes. How this is integrated into remote asynchronous replication recovery is at the moment unclear, especially with respect to latency and its impact on locking rates; For example, some type of arbitration will be required to resolve conflicts between the two nodes.
Stage 4: Global Global VPLEX introduces the ability to have multiple VPLEX nodes all connected by fibre channel links to create a global transparent network. This is EMC's federated storage vision. The vision is compelling, but the details are unclear at this time. The history of distributed data management and locking technologies is mixed as many attempts have either failed from resilience issues (i.e. recovering large amounts of data over the network quickly) or only applied in narrow use cases (e.g. small files with limited locking). It would not be surprising if this technology will contribute more value value for file-based applications than for heavy-duty database block-based applications.
Where Does VPLEX Fit?
VPLEX is currently a fibre channel-only solution. The most compelling initial use case in our view is in commissioning and decommissioning arrays and optimizing storage at synchronous/metro distance (i.e. moving data around the network dynamically). In adddition, VPLEX enables non-disruptive migrations across generations of arrays at synchronous distance. This capability is unique with the exception of Hitachi's High Availability Manager. However, VPLEX Metro is more flexible in our view, than Hitachi's HAM in this regard because VPLEX is an active:active architecture, meaning control of data can be dynamically moved at any time across the network.
VPLEX Local is currently less mature functionally when compared with more established virtualization engines such as IBM's SVC or Hitachi's USPV. VPLEX does not support thin provisioning, thin copying, “thinning” of data or space-efficient virtual copying/cloning of data. These functions are left to be provided at the array level, which means that both VPLEX and array management functions will need to be invoked to use these capabilities, increasing management complexity. Another tradeoff of this approach is seen during replication. For example, when VPLEX replicates a thinly provisioned volume, the replicated copy must be fully hydrated (i.e. "Fattened") because VPLEX creates and controls the copy.
VPLEX Metro provides a very simple topology for true synchronous copies of data and simplification of procedures. This has the potential to be much less expensive and easier to manage than traditional array-based synchronous replications and we believe could provide much lower-cost replication solutions for a larger number of applications. However, SRDF and other similar technologies from Hitachi and IBM are well established and battle-hardened and should remain the primary topology for mission critical applications requiring agressive RPO and RTO objectives.
VPLEX Asynchronous takes the concept of VPLEX metro and extends it to asynchronous distances. This capability will compete against products from virtualization vendors (e.g. IBM, 3PAR, NetApp, LSI, etc.) that create space-efficient copies of data and transmit just the changes over the network to another site. However, comparatively, VPLEX Asynchronous is an active:active architecture and can be used as a basis for more sophisticated replication. Specifically, the unique benefit of VPLEX is a second remote copy is logically in synch with the first copy which enables higher data quality.
VPLEX Global is the ultimate federated storage topology, providing the potential for recovery across multiple data centers and flexible migration of data and applications across a transparent global network. Users should understand that VPLEX Asynch and Global are future capabilities and EMC needs to introduce products to exploit this architecture.
Is VPLEX Unique?
VPLEX is a distributed, active: active, cache-coherent, federated storage architecture which in and of itself is not unique. What makes VPLEX different is this capability has been introduced for block-based storage. This means the world's transactional systems can potentially take advantage of data at a distance, DaaD.
Will VPLEX Work?
Yes. There is a high probability that VPLEX will work at metro distances. VPLEX over asynch distance is much more challenging because resilience at distance is always hard (i.e. if something goes wrong recovery is very complicated and time consuming). Performance of and resilience of systems in lock at distance is non-trivial. The reason these things are so difficult ties to the laws of physics - if something goes wrong at long distance, copying across takes days. If response time degrades at a distance, the fix is usually very complicated.
EMC's challenge is delivering an actual solution at distance that is robust, enterprise ready and appropriate for all workloads. VPLEX as announced is a very interesting architecture with great potential. It will find pockets of success in the near-term, the "follow-the-sun" vision will take time to unfold.
EMC's VPLEX technology and federated storage vision is big and bold. VPLEX has great potential to transcend the value proposition of traditional virtualization solutions and exploid data at distance. In the near term, however, users should focus on deriving immediate benefits from VPLEX in use cases such as commissioning and decommissioning arrays, where the expense of migration can often exceed $50,000. VPLEX is evolving and Users should watch the VPLEX roadmap unfold and track its progress carefully.
The overall vision for EMC's federated storage is compelling. Obviously VPLEX will not be able to move a mission-critical large database application with high levels of I/O and locking across a continent. However, there will be many simpler environments where remote migration of data and applications will be of great help operationally. Users should start with simple workloads, gain experience with and confidence in VPLEX and apply it more fully as it evolves.
Action Item: The VPLEX architecture is mind-opening. It is revolutionary, thought-leading, bold and imaginative. Users should recognize however that only VPLEX Local and Metro are currently productized. VPLEX Asynchronous and Global are architectural blueprints that will roll out slowly. Users should ultimately judge VPLEX on the products and their corresponding business value and not the architeture alone.