02 Jul 2014 by rayheffer
Updated (July 3rd 2014): VMware doesn’t currently maintain an official Horizon 6 or View Configuration Maximums document, unlike the vSphere configuration maximums PDF which is has existed for many years. There are some maximum configurations in the Horizon 6 official documentation, and release notes but this does not include everything. Before diving into the configuration maximums below, let me first explain how they are derived. Many of the configuration maximums come from testing by developers at VMware and part of the Quality Assurance (QA) process before a product is released to GA (General Availability). These are what you’ll find in the product release notes or architecture guides. Other configuration maximums are derived from best practices, benchmark tests, and observations from many implementations and tests in the real world.
A configuration maximum that has caused some confusion over recent years in the number of linked clone virtual machines per datastore/LUN. For VMware View 5.1 and prior this was documented as a maximum of 64 linked clones per LUN (VMFS). Since View 5.2 this increased to 128 linked clones per VMFS datastore, and 140 if VAAI is enabled. Today with Horizon 6 this guidance hasn’t changed and this is reflected in the maximums below. If you refer to the Storage Sizing Guide for Windows 7 for View 5.2, page 6, you will see it is also stated there (in case you doubt!).
I don’t want to go off-topic as this could easily turn into a storage discussion, but consider NFS for a moment. It’s a file-based (as opposed to block) storage system and with View Composer for Array Integration (VCAI) support and many new storage vendors adopting NFS (E.g. Atlantis Computing, Nutanix, Tintri) it is becoming a very popular choice for vSphere and Horizon deployments. With that in mind, we don’t suffer any limitations on maximum linked clones per NFS datastore, although best practice would suggest 180-250 as a balanced and tested maximum.
|Hosts per cluster||32||Max hosts per cluster supported with View|
|Linked clones per VMFS LUN||128||Best practice maximum|
|Linked clones per VMFS LUN (VAAI)||140||Best practice maximum|
|Linked clones per NFS||180-250||Best practice maximum|
|Virtual desktops per pool||2000||Maximum number of virtual machines in a single desktop pool|
|Virtual desktops recomposed at a time||2000||Based on VMware documentation and testing|
|Replica datastores per desktop pool||1||Hard limit|
|Active Connection Servers||7||Supported maximum of active View Connection Servers per View instance (Pod)|
|Simultaneous (Direct PCoIP/RDP) connections per Connection Server||2000||Based on VMware documentation and testing|
|Simultaneous (Tunnelled RDP) connections per Connection Server||2000||Hard limit|
|Simultaneous (Tunnelled PCoIP) connections per Connection Server||2000||PCoIP Secure Gateway sessions|
|Simultaneous (Blast/HTML Access) connections per Blast Gateway||800||Recommended maximum (previously 350)|
|Virtual desktops per vCenter||10000||Maximum supported (*2000 recommended)|
|Active Sessions per View instance (Pod)||10000||7 Connection Server|
|Sites per Cloud Pod||2||Maximum supported sites per Cloud Pod|
|View Pods per Cloud Pod||4||Maximum supported View Pods per Cloud Pod|
|Active Sessions per View Cloud Pod||20000||Supported maximum active sessions in federated Cloud Pod|
|Virtual SAN hosts per cluster||32||Supported maximum (**16 recommended with Horizon deployments)|
* Whilst a single vCenter for a View Pod is supported (10,000 desktops) this isn’t recommended. Each View block (delimited by vCenter) should be sized at 2,000 virtual desktops per vCenter. This avoids a single point of failure and distributes maintenance and provisioning operations across vCenter servers (E.g. power-on, recompose, refresh)
** Virtual SAN presents a single datastore per cluster. Whilst a 32 host cluster is supported, vSphere HA supports 2048 powered-on virtual machines per datastore which would equate to 64 virtual desktops per host. For better VM:host ratio and consolidation 16 host clusters are recommended.
Comments are closed for this post.