Up, in and out….Scale Out
SDX appliances now support clustering with 10.1 release.
In the other ‘Up, In & Out’ posts I’ve covered Scale Up and Scale In as part of NetScaler TriScale. Here’s an overview Scale Out which is available with a Cluster Licence on the MPX & VPX appliances.
Scaling upMPX appliances has its limitations due to the actual hardware. Once an MPX7500 (1GB) has been upgraded to an MPX9500 (3GB) there’s no where else to go as 3Gb of throughput is the maximum capacity of the tin. Each of the physical appliances shared across the range has computational and resource limitations to enable the pricing to remain competitive. If an organisation needs more than 3GB in this example then the only option is a rip out and replace with a bigger appliance or the ‘forklift’ upgrade as it’s affectionately known.
Version 10 firmware which was released earlier in the year offers TriScale Clustering – the ability to cluster up to 32 NetScaler appliances together to work as one. In the smallest example, 2 VPX10 appliances could be clustered to offer a 20Mbps capacity or at the other end of the scale, 32 of the big MPX21550s could happily manage a combined total of a touch over one and half terabytes of throughput from a single management console. The use cases between the two examples are many and varied but there’s also the ever-present high availability (HA) pair ‘elephant in the room’ to consider.
After saving for months or committing to high monthly finance payments it’s hard to imagine a customer completing the paperwork at a BMW dealership and being presented with their shiny new pride and joy along with another identical one for when the first one breaks down but this is exactly what thousands of organisations have with their IT infrastructure and the HA pair. A production appliance to look after things and another one sitting there ready to take over but actually doing nothing. IT savvy Finance Directors hate HA pairs. The money has to be spent for infrequent benefit so applying a cluster licence to a HA pair in current production would allow greater capacity for peak/growth traffic but still maintain a service in times of an appliance going offline.
Feedback from the frontline however favors the ability to run a number of smaller appliances in a cluster with a single backup appliance (n+1 config). When traffic volumes are small, fewer appliances can be applied to the stack but when demand dictates, more appliances can be added on the fly to increase capacity and maintain SLAs.
Have a look at the TriScale Clustering post for more use cases- Al
© Al Taylor 26th Sept 2012