ESG Senior Lab Analyst Kerry Dolan discusses ESG's testing of the Datrium DVX, a combined software and hardware solution that let’s you use server-supported, host-side flash to deliver application performance and data services like RAID.
Announcer: The following is an ESG Lab video.
Kerry: Hi. I'm Kerry Dolan with a summary of our recent ESG Lab testing of the Datrium DVX. Server virtualization can deliver huge resource and TCO benefits. But getting the right VM performance isn't easy, and overprovisioning resources to solve that problem is just expensive and wasteful. According to ESG research, the key storage challenges for VM environments revolve around cost and performance, storage and staff time needed to maintain VM performance, properly sizing the storage, bandwidth limitations for workload spikes, handling various SLAs, etc.
Datrium DVX is a combined software and hardware solution that lets you use any server-supported, host-side flash to deliver application performance and data services like RAID. Dedupe and compression help create a massive read cache, so all reads are local, and as a result they're fast. Rights are stored on the Datrium NetShelf, a highly available, shared disk appliance, and are deduplicated post-process. Most important, you manage VM application performance within vCenter using the DVX plugin, instead of managing storage. So if a VM needs better performance, you can VMotion it to another host with more resources.
ESG Lab testing of the Datrium DVX included fast and easy installation, plus exploration of the dashboards in vCenter that monitor IOPS, throughput, end-to-end application latency, cache hits which are normally 100%, and much more. Key performance metrics and charts show what's happening at a high level, as well as with specific hosts in the NetShelf. We also tested several ways to manage performance, including adding hosts, choosing hosts with more resources, and using what Datrium calls "Insane mode".
With the Datrium DVX architecture, performance increases as you add hosts, since each new host brings CPU and flash with it. The opposite happens with traditional network storage. More hosts means degraded performance as storage resources are spread thinner. We validated simulated workload testing that clearly demonstrated linear performance improvement as hosts were added. This OLTP test shows that as we scaled from one to four hosts, IOPS increased in a linear fashion, while latency remained low.
Datrium DVX can run in Fast mode, which reserves up to 20% of hosts' CPU for DVX processing, or Insane mode, which leverages up to 40% if that CPU is available. This can be configured on the fly without interrupting any processes. In our test setup, with 4K and 32K random read IOs, Insane mode made a big difference, delivering 49% to 200% performance improvement depending on the number of CPU cores.
So why does all this matter? The ability to mix and match host-side resources to meet the needs of application workloads provides a level of flexibility not typically found in array-based or hyper-converged infrastructure solutions. Managing VM performance is much simpler using vCenter, instead of manipulating networks or suffering through a storage controller upgrade. With traditional solutions, high and low-priority workloads compete equally for resources. But with Datrium, you can easily move a VM to avoid its noisy neighbor.
When hosts are added, Datrium DVX actually increases performance, instead of diluting it like traditional network storage. Finally, costs can be reduced due to greater resource efficiency and server-side flash. For more detail, please download the full report at datrium.com.