Citrix TriScale – scale-up, scale-out and scale-in – is about elasticity, flexibility and investment protection. It might seem that scaling-out and scaling-in are total opposites. If I were to do both simultaneously, they’d cancel each other out.
With NetScaler SDX and TriScale clustering, we can in fact do both, and get way ahead.
- Scale-out is all about building N+1 clusters. Many devices, acting as one.
- Scale-in is all about simplifying virtualizing and then consolidating network infrastructure, but without compromising on the SLAs and flexibility we expect from dedicated appliances.
The key is that with NetScaler SDX and TriScale clustering, it’s not the devices that are clustered, but rather the instances on the devices. And just like a single instance on an SDX is independent and isolated from all others, a cluster of instances striped across multiple SDX appliances is independent and isolated from other instances or clusters on the same SDX appliances. The cost and high availability efficacy benefits of TriScale clustering are amplified by the virtualization and consolidation benefits of NetScaler SDX.
Let’s use a simple scenario as an example. Lets say I want:
- 10 instances
- each instance to have three CPU cores of processing power
- the ability to keep running at full capacity even if an entire device fails
Doing this with dedicated appliances requires 20 appliances configured into 10 HA pairs. Fifty percent of my capacity isn’t used. Ugh.
With NetScaler SDX and TriScale clustering, I only need four appliances, and only 25% of my capacity is reserved for HA. Here’s how it’s done:
- I create 10 clusters
- Each cluster is composed of four instances, with each instance allocated a single core
- Each cluster is “striped” across all four SDX appliances
With all four appliances running, each cluster has four CPU cores of processing capacity. If an appliance goes offline, then each cluster still has three CPU cores. I can of course continue to scale the clusters out by adding additional SDX appliances, creating instances on the additional appliances and then adding those instances to the cluster.
All the “single-system image” benefits of TriScale clustering apply to each cluster.
- Traffic to single VIP can be processed across all nodes of the cluster.
- the cluster is configured and managed from one place
- Policy is applied to the cluster, not node by node
And, the clusters are independent of each other. I can have one cluster running one version of NetScalerOS, and another cluster running a different NetScalerOS version. If I need to do maintenance on one cluster, that maintenance doesn’t impact any of the other clusters.
So I am scaling-out and scaling-in at the same time, but amplifying the benefits of each