There was a great write-up posted on the Citrix community blog site last week that focused on improving the performance of Provisioning Services (PVS) using CIFS shares. In it, the author provided keen insight into the inner-workings of network shares and how PVS works with network shares by default. The blog was not only informative, but very helpful as it also provided suggestions on how to improve the performance of PVS, with explicit directions on what changes need to be made within the registries of PVS Servers in order to achieve better performance. And while the information provided by the author should prove very useful to PVS customers, there are several aspects that warrant further review.
Allow me to start by saying I am in complete agreement with the author that having to create and present multiple LUNs on SAN storage in order to host separate PVS data stores on multiple PVS Servers introduces additional management overhead and complexity as every time a change is made to the contents of a vDisk, that vDisk must be copied to each LUN in order to ensure vDisk version consistency throughout the PVS Farm.
I also agree that the use of “Read Only LUNs” is not practical in production environments as it presents additional management overhead and complexity in the form of multiple LUNs, replication and the use of a separate management wizard to toggle the LUN from Read-Only mode to Read/Write mode and back to Read-Only mode whenever changes to a vDisk are made.
Finally, for customers that do not have SAN storage, a network share allows a single storage repository to be used to host the PVS data store, thereby simplifying the management and maintenance of vDisks, while allowing vDisk streaming to be load-balanced amongst the servers in the PVS Farm.
But for those organizations with investments in SAN storage, employing the use of a CIFS share to host their PVS data stores would most likely be viewed as counter-productive. These customers have already realized the inherent benefits of SAN storage, including scalability, reliability and high performance via dedicated, high-bandwidth connections and block-based access to data, and would most certainly want to continue leveraging their investments in SAN storage to achieve all of these benefits for their PVS implementations as well.
And it’s not just about getting the most out of their investments in SAN storage, it’s also about maximizing the return on their investments in Citrix solutions. For example, referencing the test environment mentioned in last week’s blog, only six target devices were used. While I understand the point of the test was to validate the concept that enabling the use of oplocks and SMB could produce significant improvements in performance, the problem is that as the number of PVS Servers increases to support the growing number of target devices (which for XenDesktop and XenApp customers is just a matter of time), the likelihood for devices to encounter performance degradation also increases as a result of the locking contention often experienced in mid- to large-size file-sharing (CIFS) environments that depend on network opportunistic locking (a single, higher-layer locking mechanism). Although it is unlikely this problem would be encountered using a small number of devices, it is not uncommon for most customers to begin their initial PVS deployments using several hundred devices, with plans to double or triple the number of devices within several months. Many customers plan on scaling to well over a thousand devices within as little as a year’s timeframe!
Because of the greater potential for locking contention to occur with CIFS as I/O demands increase in conjunction with an expanding environment, not only will customers encounter performance degradation that adversely affects user productivity, the ability to scale out their PVS Farm will also be severely hindered, making it extremely difficult, if not impossible to extend the benefits of XenDesktop and XenApp to a larger percentage of users in order to realize the greatest return on their investments in these technologies.
To learn more about how customers are getting around the limitations associated with CIFS in order to realize the greatest return on their investments in Citrix solutions, click here: http://blog.sanbolic.com/?p=751