George Crump, Senior Analyst

To provide users a similar performance experience the traditional best practice is to use fully allocated (thick provisioned) virtual disks instead of more space-efficient storage technologies like dynamically allocated (thin provisioned) disks or dynamically allocated snapshots (clones). The reason for using fully allocated disks is that the more space efficient technologies, when delivered as part of the hypervisor, take a significant toll on overall storage performance.


This loss in storage performance is caused when a write occurs to the virtual machine’s disk through the hypervisor. Prior to processing the actual write the hypervisor must, on the fly, allocate space to the disk partition as well as writing the actual data. Multiply this activity across dozens of I/Os per desktop in steady state and thousands of virtual desktops and it becomes apparent why the hypervisor is a significant bottleneck. The situation is worse in spike use cases like morning boot up, login, virus scans and application updates. In these spike situations the required IOPS can overwhelm even enterprise storage systems.


To regain storage efficiency without sacrificing performance many organizations will choose a high end storage system that can handle realtime allocation functions without a performance loss. Many such systems actually have processors dedicated to these allocation tasks.


As stated in the prior article this throws the cost of storage vs. the potential ROI of the project out of balance. Essentially, the need for performance in one way or another dooms the VDI project and is the reason that so many projects end up stalled or only half completed. Customers are looking for ways to address VDI storage performance issues without using pre-allocated storage volumes or buying expensive high end storage systems. They need to get the ROI for these VDI projects back in balance. The potential alternative is to look for third party storage management software like Virsto VDI that can provide performance improvements with mid-range storage solutions.



The Software Alternative


Software based storage solutions can provide performance enhancement to the hypervisor starting with the ability to use dynamically allocated volumes without impacting performance or requiring the use of high-end storage systems. This brings cost effective capacity utilization to VDI and allows the use of mid-range storage.


In the Virsto case they replace the standard hypervisor virtual hard disk with what is called a Virsto vDisk. The vDisk is a journaled volume that uses an optimized log space, which queues up writes so they can be sequentialized instead of randomly written to disk. This makes the “allocate then write” process described above significantly more efficient and perform as well as fully allocated volumes.


Enabling dynamically allocated volumes is more than just lowering the physical capacity required for a VDI environment. It also enables more freely the use of snapshot technology. In many environments snapshots can only be used on a very limited basis because of performance concerns. Most storage managers think of snapshots as just another backup. This is because they see snapshots, if they use them, as only viable for the most recent version of a desktop. When an unlimited number can be used without performance impact new possibilities open up.


  1. BulletRoll back images for before and after applying patches

  2. BulletInstant recovery points for a failed work station or corrupted application

  3. BulletRapid provisioning of additional virtual desktops based on already configured desktop

  4. BulletCreating test images when deploying new software

  5. BulletHistorical point in time recovery of files or desktops, not just the last copy

  6. BulletSnapshots are the key foundation for space efficient cloning of desktop images



Faster and Longer Lasting SSD


An important benefit of a journaled, log based system that sequentializes writes is that it can significantly enhance solid state disk (SSD) performance. Thanks to more affordable price points, flash based SSD is becoming more commonplace in the VDI environment. This is because it’s an ideal solution for dealing with boot storms and other activities where hundreds of desktops need to read and write data to the storage device at the same time. As is the case with enterprise hard drive systems, VDI puts an important weighting on the cost of those SSD drives which is particularly challenging given their already premium prices. This makes efficient use of more expensive SSD capacity even more important.


As a result VDI environments are looking to use standard, off the shelf SSD systems and to derive maximum performance as well as capacity utilization from them. A log based system that sequentializes writes and reduces the randomness of write I/O is an ideal match for flash based SSD. SSDs are an ideal storage device to utilize thinly provisioned and cloned volumes as well.


The weakness of flash based SSD is in its handling of high numbers of random write I/O traffic, but the technology is generally very effective at handling sequential write traffic. This is because data must be written to a flash device in a specific sized segment. If the data is smaller than the block size that flash device requires, existing ‘good’ data must also be read and rewritten along with the new data.


Not only does this cause a performance loss - it also reduces the life of the drive since flash memory SSDs have a finite life span of anywhere from 10,000 to 100,000 write cycles, depending on the type of flash used in the drive. Rewriting data to match a particular block size simply means that unneeded writes are occurring and are reducing the lifespan of the drive.


The sequentialized writes of a log based system address both of the problems, because new writes are grouped together to produce larger net-new segments, which are less random. As a result, a log device built from SSD operates at “sequential write” performance levels rather than the “random write” performance levels that are typical. These larger blocks also mean less rewrite of existing data to “fill-up” blocks of flash memory, which in turn increases the lifespan of the SSD.



Performance Management Through Tiering


Finally a system like Virsto’s also brings a tiering structure to the VDI environment similar to what high-end storage arrays typically offer, but again in mid-range systems. Tiering allows the storage manager to make sure that clones or golden master images of desktops can exist on an SSD-based tier 0 and that more performance-neutral user files be accessed on less expensive hard disk based storage. A software based storage solution makes this a reality.


If the environment has to use fully allocated volumes for the storage of VDI images then using SSD to store those images would be impractical from a cost perspective. Golden masters (clones) are essential to making SSDs a viable option for VDI but those dynamically allocated volumes need to perform at an acceptable level. Using a software based storage management service like Virsto also makes that possible.


With that foundational capability data can be placed, cost effectively, on the most performance-appropriate tier. Virsto for example, provides that capability through the GUI and allows for defining which tier should hold which type of data, greatly simplifying the storage provisioning process.


A software based system that brings dynamic allocation functions like thin provisioning, snapshots and clones to storage hardware provides greater flexibility when choosing a solid state storage system. Many of these systems don’t have the advanced features of enterprise systems. Since products like Virsto are software based the ability to mix and match hardware to a specific storage need, even within the VDI architecture, brings great flexibility and further cost reduction.



Summary


VDI presents an unusual challenge to the storage manager. First, capacity must be very effectively used with dynamically allocated volumes, which keeps costs down and provides features like thin provisioning, snapshots and clones. Second, consistent, predictable performance must be maintained so that the end user’s experience is at least as good as it was with their standalone desktop. This is something that built-in thinly-provisioned volumes from the hypervisor companies can put in jeopardy. Enterprise-level systems can provide these capabilities but at a cost that often ruins the project’s ROI.


Software that uses log-based journaling as does Virsto, may be the ideal alternative. It can offer the thin provisioning, snapshot and cloning capabilities of enterprise systems but do so on affordable mid-ranged storage, which can be mixed and matched according to purpose.

Virsto is a client of Storage Switzerland

- Performance