A Hybrid Storage Array Comparison

Choosing storage:

In our environment, our requirements were a mix of price, space, and much greater performance than our traditional, spinning disk-based, iSCSI system we were operating with and certain workloads, such as our Cisco VoIP VMs.

For four years ago, your general classes of SAN storage would have been defined by connectivity only, Fibre Channel, NFS, or iSCSI. Today we are seeing this as a secondary level of options, with the prevalence of flash defining the first level.

• Top-tier storage devices are called all flash—arrays loaded with nothing but high-end SSDs. These devices are seen from vendors such as Pure Storage and SolidFire.
• Next, you have what are known as hybrid arrays—those with a mix of flash and traditional spinning disk used in various methods of caching.
• Finally, you still have the old, stoic traditional storage systems, loaded with traditional spinning disks, filling the racks from head to toe in your datacenter.

For vendors dealing with hybrid arrays, those that used a mix of flash storage and traditional spinning disks to provide a high number of IOPS. And we looked hard at the technology, including going booth to booth at VMworld and reviewing vendors, and we narrowed our search to NetApp’s FAS series devices, Nimble Storage, and Tintri Storage devices.

NetApp:

It is a very powerful, and surprisingly cost effective, system once the negotiations are done for the base system. The combination of hardware and NetApp’s software stack is legit and allows for a great deal of flexibility. So connectivity can be any that you want, but my impression is that most is done via NFS.

Although the software stack is powerful if you really only need an array or an array with a bolt on the shelf, it is a bit intimidating for the IT generalist when compared to the other options.

Tintri:

It is the Cadillac of the group in terms of speed of deployment and ease of management. Tintri’s appliance model means that, when you fill up a device, you can’t just attach a relatively less expensive shelf to your existing array; you have to drop in an entire new unit.

The claim is that you can rack an array, plug it in, configure networking, and be deploying VMs to it in under 15 minutes. Again, connectivity is done via NFS.

Tintri is there are no LUNs to configure; the entire device just presents as a giant single data store. After the data goes to disk, it is reinflated. Also, it is great if you are a Web 2.0 type of environment.

Nimble Storage:

Nimble really gave the warm and fuzzies. Management and deployment were painless; configure a management VLAN and a data VLAN, plug it up, add an IP, and go. Layer in the vCenter Plug-In, and you can then create and deploy your data stores directly from either vCenter client, with LUN creation, performance tuning, and the ability to add the data store to any and all hosts—all done in a simple-to-use wizard.

Deduplication and compression are done all the way through, and we have seen some great results. And our VM environment is fully configured and running with thick provisioned disks. After converting the majority of VMs to thin provisioning and having dedupe and compression applied, those same VMs are currently using 2.6 TB in total.

Anyway, these new grades of storage arrays provide greater performance for competitive prices and all will provide good, easy-to-understand insight into the performance and monitoring of the systems.