Premise
EMC has announced (on 2/29/2016) an all-flash array (AFA) version of the Tier-1 VMAX, the VMAX AFA. This allows a potential IO performance improvement of an order of magnitude, from a 5-10 millisecond IO performance level with existing EMC hybrid VMAX solutions to a sustained IO performance level in the 300-500 microsecond range. Enterprises having high-value applications with well established high availability VMAX infrastructure can now avoid expensive migrations to all-flash arrays just for performance reasons. Enterprise VMAX hybrid storage arrays with extensive use of Oracle database have other solutions that can extend the life of existing VMAX hybrids.
Wikibon’s assessment is that the EMC VMAX AFA is a good tactical step on the digitization journey, and will need significant investment and improvement by EMC/Dell before it can claim to be a strategic path.
Definition of Tier-1 Storage
Wikibon has always defined tier-1 storage as meeting three key criteria:
- Has the time-tested ability to throw whatever array resources are required to allow critical applications and the underlying databases perform at the highest possible levels required by the business;
- Has a complete and fire-tested set of high availability remote replication functionality allowing large-scale consistent replication and recovery to/from multiple locations;
- Has a set of well-established performance and availability services that deeply understands both the technologies and how to integrate and project manage those technologies to meet the specific organization and industry application imperatives, and prove those requirements have been met.
EMC has been able to keep pace with requirement 2 & 3 with the hybrid VMAX, but has not been able to keep pace with requirement 1 for Tier-1 storage performance when compared with all-flash arrays.
– VMAX All-Flash Availability
The advantage of the VMAX is the deep technical and business understanding of high availability, and the ability to put together configurations, processes and procedures that “take a lickin’, and keep on ticking’ “. Wikibon has consistently shown that the cost of migration of arrays is high and the ROI on migration projects low, as in its recent research on NetApp array conversion. The VMAX AFA significantly lowers the high cost of migration from hybrid VMAX just for performance reasons.
Another different approach is discussed below in the section “Alternative Approach for Oracle-heavy VMAX Hybrid Environments”. This approach maintains the VMAX availability and change control procedures, minimizes migration costs, and can extend the life of the VMAX hybrid arrays.
– VMAX All-Flash Performance
The flash performance available on the VMAX all-flash array is much better than the previous VMAX arrays. It is also better than AFAs which have built-in de-duplication and compression, e.g., around 900 microseconds. However, it is still not best of breed. The use off SSDs still holds back the highest IO performance, as does the lack of NVMe native support.
EMC’s DSSD GA announcement on the same day is an example of much higher performance, using PCIe-based flash modules, PCIe switches and NVMe protocols to provide another order of magnitude reduction in IO latency.
Wikibon expects that future EMC all-flash VMAX “V-Bricks” will introduce more advanced architectures and higher performance.
– VMAX All-Flash Services
EMC has earned the long-term trust of its customers when its comes to supporting their customers, who have come to expect that when a serious malfunction occurs, the skies will darken as technical experts are parachuted in. One of the challenges of a merger with Dell is to maintain the confidence of EMC customers in future EMC/Dell service levels for Tier-1 computing.
– Alternative Approach for Oracle-heavy VMAX Hybrid Environments
For Oracle-heavy VMAX environments, an alternative approach to replacing hybrid VMAXs with VMAX AFAs is to take advantage of the Oracle ASM software, and use a very low latency ASM-mirrored AFA to offload the read IOs. HPE has tested such an approach using the 3PAR AFA, and IBM, Oracle and Violin have similar very low-latency AFAs. This approach maintains the VMAX high availability infrastructure and procedures, significantly avoids migration costs, and can extend the life of the VMAX hybrid arrays.
Another approach is to deploy EMC’s DSSD for Oracle-heavy environments, either standalone or more likely as an ASM mirror as discussed above. The use of VMAX with excellent write capabilities and data services, and using preferred read to push all reads to the DSSD combination of servers and extremely low latency flash would be the most performant and cost effective of running large performance constrained Oracle workloads. EMC have done some extensive testing with Oracle on the DSSD, but would need to perform extra testing and certification in this configuration. For these IO constrained environments, the additional system throughput provided by the much lower latency of DSSD could be easily justified by reduced Oracle licensing costs and reduced programmer/DBA maintenance costs. The DSSD can also become a component in the EMC VCE Scale architecture, and could in future be part of a converged infrastructure.
Strategic Implications
The introduction of the EMC VMAX AFA has signaled EMC’s focus on enterprise all-flash products, and a clear role in the EMC/Dell merger.
EMC are including the VMAX AFA V-Brick as a building in its Vblock and Vscale range of converged infrastructure offerings. This is an important option for creating converged “True Private Clouds”, and moving the $300 Billion spent worldwide on enterprise operational personal into vendor R&D, converged and hyper-converged systems.
The VMAX AFA does not have the cost efficiency from data reduction that can be expected from other AFAs, such as EMC’s XtremIO or Pure’s AFAs, but has better IO performance. Compression will be available in a future release. For workloads with less stringent RPO and RTO requirements, there are many lower-cost alternatives.
The key strategic imperative for enterprise CXOs is to be in a position to take advantage of the Systems of Intelligence, and enable data driven automation. Wikibon lays out a roadmap to Systems of Intelligence in its research entitled “Ten Strategic Steps for Meeting the Data Driven Imperative“. In order to reduce “Time-to-decision” from the current days to sub-second will need a whole-scale change in the way data is managed and technologies and software are deployed. The EMC VMAX AFA is a tactical step on the journey, and will need significant investment and improvement before it can claim to a strategic path.
Action Item:
For senior executives with significant investment in tier-1 VMAX infrastructure and its high-availability eco-system, the announcement of the VMAX AFA is a welcome relief, and will mean a significant short-term cost and risk avoidance of migrating to other AFAs just for performance reasons. The VMAX should definitely be included on all RFPs in all performance constrained VMAX environments. In addition, senior executives with significant Oracle database on VMAX should ask for EMC to test and bid Oracle database ASM based solutions, using low latency mirrors. They should extend the RFPs to low latency vendors who have tested and can provide such solutions.
However, moving to an all-flash data center is more that just keeping the status quo of existing methods for managing data. Senior executives will need to map out a clear strategic infrastructure path to ensure the enterprise can take advantage of full digitization and Systems of Intelligence.
