A closer look at Kaminario

kaminarioAs many of you are aware, I was at VMworld in San Francisco last week. I wrote a number of articles about some VMware storage announcements, such as EVO:RAIL, VAIO and VVols. However there were, as usual, quite a number of storage vendors at this years conference. One of the vendors that I really want to learn more about was Kaminario, an all flash array vendor that I’d heard a lot of things about. I had the pleasure of spending some time at the Kaminario booth with Shai Maskit who is a senior Product Manager with Kaminario. I posed my usual set of questions to learn a bit more about their AFA products.

The K2 All Flash Array is built out of K-Blocks. Each K-Block comes as a dual controller, 4U system. The controllers run in an Active/Active configuration. The K-Block also houses 24 Solid State Disks (SSDs). The SSDs are enterprise class SSDs, not consumer grade. There are two supported SSD sizes at the moment, 400GB or 800GB. All components, such as SSD, battery packs (for destaging in the event of a failure) and power supplies are hot swappable.

Scale Up or Scale Out

Both! Shai stated that the K2 can be scaled up and scaled out. For Scale Up, a second SSD shelf containing another 24 SSDs may be added. For Scale Out, up to 8 x K-Blocks can be added to the configuration, and each of these could also contain an additional shelf of SSD. The interconnect is point-to-point infiniband. In order to achieve scale out, an infiniband switch is required, but a single switch can support the scale out to 8 K-Block systems. Two infiniband switches are used for high availability. A single stand-alone K block does not require the switch. Shai also stated that scale out is completely non-disruptive to the already running workloads.


Currently the K2 supports iSCSI and Fiber Channel. Each controller has two ports, providing a total of 4 ports for connectivity from each K-Block.

Data Services

Thin Provisioning, inline compression and inline global deduplication (across all the K-Blocks) are all available as data services on the K2. The deduplication is optional and can be enabled at the per-LUN level. This avoids deduping applications that may not benefit from it, for example, a database application. Shai stated that their customers normally enjoy a 6:1 data reduction ratio (dedupe and compression combined). Some customers have had even higher data reduction ration, but Kaminario will guarantee at least a 3:1 ratio. The  data reduction ratio is visible from the K2 dashboard, alongside other performance and capacity information:

K2 GUI dashboardKaminario also offer encryption on data at rest, should you have such a requirement.

From a snapshot perspective, snapshots on the K2 can be nested and can go to a depth of four. Kaminario also offers consistency group snapshots for applications that may span multiple LUNs (snapshot multiple LUNs at the same time).

Kaminario do not offer replication at the moment, but this is in development and they plan to provide asynchronous replication by the end of this year.

There is also no specific Quality of Service mechanism for I/O performance. There is however QoS on capacity, which is also essential when it comes to guaranteeing capacity per groups of volume and their snapshots. This is especially true when you consider the various data reduction features available, such as thin provisioning, inline deduplication and inline compression. Shai went on to say that the K2 performance avoids noisy-neighbour and SLA per volume issues. However, he did go on to say that this could become an interesting data service in the world of Virtual Volumes, so QoS of performance is something they are definitely looking at.

vSphere Integration

The K-Block supports the full set of VAAI primitives for block storage – that is, ATS, XCOPY, ZERO, Thin Provisioning and UNMAP. They also have a web client plugin, which allows joint customers to manage multiple K2 systems and their vSphere infrastructure from a single interface.

vcenter 4There is no SRM integration as there is no native replication currently, but this is something that Shai tells me that they plan to deliver when they have a native replication feature. Since this is an All Flash Array (AFA), there is not much point in looking at SIOC or Storage DRS based on I/O metrics integration, since the latency thresholds for these vSphere features are still too high at 30/35ms, something we’ve heard from other AFA vendors in the past. However Storage DRS for initial placement and load balancing based on capacity is still something that can be utilized.

Virtual Volumes

Shai stated that Kaminario’s plans for Virtual Volumes (VVols) are aligned with the VMware road map. They are members of the VVols program, and are working on building the right solution for a storage array with a scale-out architecture.


Looking at their website, the K2 appears to come in at $2/GB. This is pretty good, but I’m not sure if this is for a basic K2 with no shelf unit, or if this is for a much larger system configuration. Another interesting point to note is that these guys came away with Best of Show from the 2014 Flash Memory Summit – “the most innovative flash memory technology”. So, all in all, this looks to be a nice AFA offering, with lots of data services and some good vSphere integration points.

  1. From the storage architecture point of view this array follows the same architecture principles as the EMC XtremIO AFA, that is scale-out to a certain extend and tightly coupled.

    And with the all-in basic set of data services and $2/GB, Kaminario can really competes with XtremIO.

Comments are closed.