Tintri upgrades virtual machine QoS and takes a shot at VVOLs

Hybrid flash array maker Tintri adds IOPS performance guarantee per VM to its operating system and takes a swipe at VMware’s VVOLs

Hybrid flash array maker Tintri has announced upgrades to its storage operating system (OS) that add the ability to assign guaranteed levels of performance to individual virtual machines (VMs).

The move shares similarities with VMware’s introduction of Virtual Volumes (VVOLs), but Tintri says its implementation is superior because its basic addressable unit of storage is more finely grained.

The ability to assign performance to individual VMs is significant because of the way virtualisation has changed how hosts (servers etc) relate to storage.

In the past, one physical server was assigned storage on one volume or LUN made up of a number of physical drives on a storage array. With virtualisation, multiple virtual servers or desktops are assigned common storage resources. This boost to utilisation is in part an advantage of virtualisation.

But, often that has meant VMs with different performance requirements have had to reside on the same LUN. Tintri’s VM-level functionality and VVOLs aim to assign virtual machines to storage resources suited to their performance needs – depending on the availability of physical media of different classes.

Read more about VM storage

  • VMware’s Virtual Volumes (VVOLs) promise a revolution in storage for virtual machines, but what are the key things you need to know? 
  • Virtualisation and the LUN: Storage administrators used to match LUNs to physical servers, but that’s all changed. Find out how in this guide to the basics of VM storage.

Addressing latency in bulk capacity

In the upgrade – to version 3.2 of its OS – Tintri has added Quality of Service (QoS) functionality that provides the ability to allocate maximum and minimum IOPS to individual VMs, as well as visualisation in the user interface (UI) to allow administrators to see the impact of changes on VM-level latency.

Tintri said the tool will enable visualisation across the entire infrastructure, including latency stemming from host, network and storage contention.

Tintri marries low-latency flash with bulk capacity on spinning disk and specifically targets virtual machine environments. To do this it does away with volumes, LUNs and Raid groups, and maps I/O requests directly to the virtual disk.

Tintri’s key selling point is that its storage is addressable at the level of the VM, and it does away with the traditional unit of storage provisioning, the LUN. It is this basic architectural principle, said Tintri product marketing director Chuck Dubuque, to which the company has added the latest QoS functionality.

“We have always had a QoS story because we treat each VM as an individual entity with performance isolation between VMs and with VMs given ‘fair share’ of performance resources available,” he said. “In this release, we are adding QoS visualisation with the ability to take VMs from the ‘fair balance’ pool and give them their own scheduled policy that kicks in when resources become overloaded.”

Avoiding contention in assigning performance

The new functionality – which can be controlled via sliders in the UI or via PowerShell and REST APIs – is similar in intent to VMware’s VVOLs, which were introduced with vSphere version 6 earlier this year.

VVOLs allows storage administrators to assign specific levels of storage performance to individual virtual machines. Array makers' support for VVOLs allows VMs to access tiers of storage with different levels of performance.

Tintri supports vSphere 6 but does not have a date for support of VVOLs, said Dubuque. He added that Tintri’s granularity of performance guarantees per VM are in any case superior to those offered by VVOLs.

“VVOLs provide a lot of the services we offer with our QoS functionality, but VVOLs still depend on finding the right level of performance based on LUN-level provisioning,” Dubuque said.

“That does nothing to address the ‘noisy neighbour’ issue where requirements for different levels of performance contend on the same LUN. We will deploy VVOLs at the VM level.” 

Read more on Virtualisation and storage