Dinesh Singh

Is All-Flash holy grail for VDI ?

Blog Post created by Dinesh Singh Employee on Oct 19, 2015

For several years, all-flash array (AFA) vendors (as defined by Gartner!) have tried to create a niche in the desktop virtualization market by touting AFAs as the most appropriate solution for VDI installations. However, just as with other innovations, all-flash arrays have followed their own adoption curve. Although price per GB has fallen over the years, flash continues to be significantly more expensive than magnetic disks. Even other associated benefits such as power and space savings haven’t been able to help build a business case for mass adoption of all-flash arrays due to prohibitive cost considerations. My colleague Bob Madaio cracked  a joke on all-flash !


Many customers perceive AFA as a good candidate for only a few transactional workloads which have homogenous IOPS and latency requirements. For the rest of their workloads, they still invest heavily in tiered storage platforms, combining flash and spindle disks to serve applications with performance requirements that may change over time. By assigning workloads to the appropriate storage tier per their performance requirements, customers can consolidate data on a single storage device for a better TCO. This is true even in virtualized environments because not all VMs need tier 1 storage services to meet IOPS and latency requirements. Hybrid storage using Dynamic tiering, active flash can perform similar to an all flash environment for less cost by using lower cost hard drives for steady state and inactive seats. In internal tests using a VDI workload and a hybrid storage configuration including hard drives and flash media with Dynamic tiering, active flash software, the average latency on storage ports peaked at less than 2 milliseconds during a login storm.


During 2015 VMworld shows, I found rather uniform pitch among many  new AFA providers that the all-flash offers consistently high IOPS performance to avoid I/O boot storms, which has been death knell for many VDI projects. An I/O boot storm is the performance degradation that usually occurs when a large number of users try to boot up within a narrow time frame overwhelming the storage array, often resulting in dissatisfied users. However, recent lab tests conducted by Hitachi engineering strongly demonstrates that well designed and workload-optimized hybrid systems can provide performance equivalent to AFAs for random I/Os and help manage surges in traffic without impacting user experience. This allows customers to implement one array that will solve VDI challenges and consolidate a broad range of workloads for improved TCO.



Hitachi data mobility delivers consistent IOPS for VDI applications during regular usage and peak workload scenarios through a new technology called active flash. Active flash is a part of Hitachi Dynamic Tiering (HDT) that improves tier 1 read/write performance by promoting pages to tier 1 during boot storm stress situations. As shown in picture 01, a hybrid system (a VSP G800 in this case) constantly monitors IO traffic and when spikes are detected, in this example a boot storm window between 0930 and 1000 am, delivering real time performance to meet the surge in I/Os when many end users simultaneously try to log into their desktop environments.


Customers evaluating storage systems for VDI implementations stand to benefit from data tiering solutions offered by hybrid platforms for performance to keep users happy and capacity to maximize ROI considerations.


Moreover, recently announced “Instant Clone” capability in vSphere 6.0 provides customers another arrow in quiver to cut down on storage costs and improve ROI from VDI projects by investing in cost-optimized hybrid-systems with active flash tiering capabilities. Instant Clone capability allows admins to rapidly clone and deploy VMs as much as 10x faster than what is currently possible today. It allows admins to fork a running virtual machine. Admins can create child VMs in second with child VMs leveraging the memory and disk of parent VM. Since child VMs are created using clones of running parent VM, they are already in full running state (when powered-on) and there is no boot up. Arguably the single biggest reason for VDI project abandonment may no longer be valid and so the justification to invest in expensive all-flash array !


What you think ? Do share your thoughts by posting comments.


Watch out this space to learn more about Hitachi storage and converged solutions for vSphere environments.