Managing storage in a virtual environment can be quite a
challenge; with constantly growing and shrinking LUNs, you have to
ensure that everything in your environment is just right. Then you have to
connect your ESX hosts and build datastores to hold any number of virtual machines
your environment requires. Whew. That seems to be quite a bit of work just to
get things prepared to function.

Fortunately, another storage vendor has taken a different
approach to the management of the virtual environment. The company is Tintri, and the approach to storage is
focused on the datastore. The VMstore appliance is a
datastore-in-a-box that aims to streamline the management of any VMware-based
virtualization environment.

Getting started

Once the unit is installed in the rack, configuration
requires little more than powering on the device and assigning an IP address.
ESX hosts will use the assigned IP address to connect to the Tintri appliance. From
here, a datastore can be created in vCenter against the Tintri VMstore
appliance. Once this is up and running, your virtual machines can be moved into
(or created against) the Tintri VMstore appliance.

Note: For the
purposes of this review, I was granted access to the Tintri’s Partner “Lightning
Lab” to get a feel for the appliance and how it is managed.

The management interface is also all web-based, removing the
need for the installation of additional software.

Figure A

Tintri Dashboard

From the dashboard the visibility of information and
operations available for virtual machines in the datastore is huge. Because of
the integration available between Tintri and vSphere using VAAI (vStorage
APIs for Array Integration
), many actions available for the guest systems,
especially cloning of guest systems, is greatly improved.

The most noticeable feature available in the application is
overall reporting and management. Almost every aspect of your environment is visible
in just a few clicks; we will look at the details of these features as we move

The dashboard displays overall performance of the datastore,
covering the following items:

  • IOPS – the number of input/output operations per
    second for the entire datastore
  • Throughput – the total throughput for the
  • Latency – the overall latency experienced for
    the datastore
  • Flash hit ratio – the percent of used flash

Figure B

Top level details for the datastore and latency information

When you mouse over a portion of the overall information,
the details that make up the collective are displayed in a tool tip. In the
demos I saw with Tintri and during the time I was able to use the lab, this
feature seemed to be one that made the most sense to me. Being able to see
which portions of an environment, guest, host, or network may be the root cause
of a problem without having to visit these pieces individually is an enormous
time saver.

In addition to the overall datastore information displayed,
Tintri looks at performance reserves and available space which presents a complete
picture of an environment. This is available in other storage platforms as
well, but only the truly usable storage is displayed. Within the Tintri
appliance, the actual resource usage for each VM is known and the real performance
data is displayed. This allows organizations to make honest decisions about
growing their storage to best suit their needs, avoiding the need to make
guesses without much consideration for the performance of the array, only the
disk space remaining.

Figure C

Performance and space

Have you ever wanted to know which VMs in your environment
are changing to consume more resources in terms of storage space consumed or
performance data? Because Tintri captures this information, they are able to break
this out by VM. Clicking on a listed VM (or using the Search VMs feature) will
show specific details about that workload.

When a virtual machine is selected, its individual
performance data (IOPS, latency, throughput, replication status) is displayed
along with information about the number of snapshots that exist for the
workload. The table shows the current information, and the graph shown at the
bottom of the page displays an ongoing collection of data. The graph can have
the timeframe modified by right-clicking the display and selecting an option
from 4 hours to 7 days.

The graph can also display loads of other information by
selecting the drop down arrow at the top to bring up different views. By
selecting CPU and Resource usage, for example, you can see time snapshots as
the mouse is moved over the graph.

Virtual hardware

When looking at a virtual machine, you can also drill into
the virtual disks used by the VM to detect any issues with the disks used by a
workload, such as the overuse of a swap file. Features like these can help
troubleshoot your virtualization environment without requiring you to log into vCenter
to investigate.

What about cloning
and snapshots?

VMware takes snapshots of VM workloads to help with backup
and recovery of the application. Tintri has considered this as well with two
types of snapshots:

  • Crash consistent – able to recover to a previous
    snapshot of a workload
  • VM consistent – uses the VMware tools to “quiesce”
    the workload when the snapshot is taken, that is, stop any running processes to
    make sure the VM version is stable

When Tintri takes a VM-consistent snap, a call is made to
vSphere and the VMware snapshot utilities are used to create the original
snapshot; then a Tintri snapshot of that is taken, and the original VMware
snapshot is deleted, leaving a compressed and storage-optimized snapshot
remaining. In addition, no resources other than a small amount of space to hold
the snapshot metadata are consumed. When a VM is cloned from the Tintri UI, a
snapshot is taken and the clone is based on that snapshot – not consuming any
space beyond a negligible amount of metadata, and the cloned VM added to
vSphere inventory. Up to 500 VMs can be cloned in a single operation in a
matter of minutes.

Additionally, through the use of a VAAI plugin, a clone can
be created and added to VCenter inventory immediately with no additional work
required. With this integration, clones created within vCenter are managed in
the same way, with the work being offloaded to the storage array for most
efficient processing – and only taking a few seconds to clone an entire VM.

Snapshots are great,
but can Tintri go offsite?

Replication is based again on snapshots; whenever a snapshot
is taken on a VM configured to replicate, the array replicates the snapshot to
a partnered Tintri array. After initial replication, only the changes to
workloads are replicated (which are globally deduped and compressed), either on
a schedule or on demand. Most arrays are able to replicate a LUN or set of LUNs
to another location, sending any information they contain to the other side. Tintri
works differently, at the VM level, allowing a single workload to be replicated
to another Tintri appliance on its own protection schedule.

Figure D

Replication configuration

When working with replication, configuration is very quick
(and on a per VM level). Simply checking a box to enable protection and
providing a schedule of replication operations is all the work that needs to
get done. As you will see below, there are some licensing costs for
replication. In addition, if you are splitting your environment into VLANs to
segregate data and want to use a distinct VLAN for replication, you may need to
discuss additional adapters with Tintri. These are not required out-of-the-box
for replication to work, but something your organization may need to consider
down the road.

Because the Tintri appliance almost completely takes over
the management of VMs within its datastore, the tools and reporting will be
hugely useful to storage administrators and virtualization administrators
alike. The device also makes available hardware diagnostics and a method to
bundle information and diagnostics for Tintri support. This information is
shown in Figure E.

Figure E

Diagnostics for

Tintri storage appliances are very useful in virtualization
environments because they simplify the management of both storage and hosts by
bringing it to a central point of administration. By allowing the storage to
handle cloning and snapshots for the virtualization environment, the execution
time for creating these items is reduced greatly. In my testing, creating both
snapshots and clones of workloads using the Tintri management tools took only
seconds to create.

Models and pricing

As of this writing, the available Tintri appliances are (prices in US dollars):

  • Model T540 – Dual Controller: 13.5TB, $90,000 (10Gbe Copper NICs) or $95,000 (10Gbe SFP + 10Gbe Copper NICs)
  • T445 – Single Controller: 8.5 TB, $65,000 (10Gbe NIC or SFP available for additional cost)

Replication is a licensed software feature that comes in
addition to the cost of the appliance with licensing for adapters as well:

  • Model T540: $16,500
  • Model T445: $11,700

At first glance, the pricing might seem prohibitive for some
organizations. However, there are other considerations. If your organization is
growing their virtualization environment and the storage it should live on,
Tintri is an appliance to consider for storage, performance, and ease of
management reasons. While the cost per gigabyte or terabyte might not be as
small as your organization would like, the performance gains and ease of
management might be factors to put Tintri back on the table.

Looking at the pricing above and seeing a bunch of big numbers
in terms of overall cost can be a bit misleading, not to say that the bottom
line cost isn’t something to consider, but looking at the cost per workload may
be a better way. When this is taken into consideration, the cost can be very competitive
with other vendors. In research for this article, I worked to determine an
approximate cost per VM in my own environment – approximately 30 VMs – (which
does not run on Tintri) and found it to be about 75$/VM. That is not a terribly
huge cost at all. In addition, in working with Tintri to lab test this
solution, they mentioned that they have a customer running 1000 VM workloads on
one single Tintri device, which brings the cost per workload down
significantly. Looking at other factors, like administrative costs to manage
storage and virtualization environments should also be worked into the
calculation of cost and cost savings. If the device takes less time and effort
to manage, it might just save money in the long run.

My hope is for Tintri to release a class of appliance that
is geared toward the SMB market, allowing more organizations to consider Tintri
for virtualization.