Couple months back I wrote an introductory blog on NetScaler Cluster (/blogs/2013/09/12/netscaler-cluster-the-way-ahead/) and always wanted to follow up with more blogs and data around this awesome technology and architecture. As we see more and more customers moving towards Cluster adoption, here is the next level data on important concepts.
NetScaler has couple important types of IP addressing schemes in the system. As a NetScaler user I am sure you deal with NSIP, VIP, MIP, SNIP on daily basis and nothing changes here. For new users of NetScaler, NSIP is the management IP of the NetScaler system, VIP is the virtual server IP, MIP is the mapped IP and SNIP is Subnet IP. We have enough documentation around these IP addresses on docs.citrix.com to read more. Cluster introduces new concepts like CLIP, Striped IP, Partially Striped IP and Spotted IP.
CLIP is the Cluster Management IP address which is owned by Cluster Master. Unlike NSIP which is pinned to a specific system, CLIP remains floating and is always associated with the Cluster Master node. So what is Cluster Master? Cluster Master is a NetScaler node which is part of the Cluster and logically owns the responsibility to manage the Cluster subsystem. Next question, who decides Cluster Master? Well the decision here is not left to any one node or admin, every node which runs the Cluster protocol takes part in the election process of Cluster Master. Thus one node gets elected as Cluster Master based on several parameters like health of the system and other node reachability etc. Every time the election happens a different node can pick up as Cluster Master. Thus it is not pinned down to one system at all and is a floating concept. Now you must be thinking how other node will pick up as Cluster Master? In our Cluster architecture, every node which is active and is processing traffic has complete details of the whole Cluster and can take up the responsibility of Cluster Master. Thus all Cluster nodes has same level of data at all points and provides much better fault tolerance as Cluster Master going down is as good as any other node of the Cluster going down. Cluster sub-system will elect a new Master as soon as the existing Master node goes down. Next question, what is Cluster Master supposed to do? Beyond participating into normal traffic processing, Cluster Master is responsible for other management tasks like configuration synchronization, configuration propagation, ensuring every other node is updated with latest. It also owns other manageability tasks like user interface, reporting, monitoring, tracing, logs etc. Hmmm… Did you notice we left the original topic of CLIP much behind :). So as you understand Cluster Master better, it is easier to explain that CLIP is always associated with Cluster Master and is the most important IP for an administrator to connect to. All administration tasks can be done through CLIP like accessing the unified GUI which ensures that configuration is propagated to all nodes. In a Cluster topology you need to ensure that individual NSIP need not be accessed for all day to day activities. Do note that Cluster Master owns the CLIP and does have its own NSIP for the node. For all kind of configuration and management tasks, one rule – always connect to CLIP.
Now let us get to the other 3 type of IP addresses which Cluster architecture introduces.
Striped IP – think of it as a logical IP available on every Cluster node thus providing scalability to the Application/vserver. Striped IP is associated with every node and is applicable to all kind of IP addresses like VIP, MIP and SNIP. Thus for true scalability and performance you are looking for deploying Striped VIP as front-end.
Partially Striped IP – as the name suggests, this is the type of IP which is deployed across some of the nodes or a group of node recognized as “node group”. If you think of practical deployments, you will run into several scenarios where you want to scale the App to more than 1 node but not to all in a 5 node Cluster. This gives you flexibility in terms of selecting what you want to scale and at which level using which nodes.
Spotted IP – by now you must have guessed that this is the one pinned down to a single node and that is correct. Spotted IP is configured only on single node and has visibility for the given node only. Spotted IPs also has great use case like using SNIP as Spotted on every node. As you know Cluster requires traffic steering to ensure every node is participating in the game and by using Spotted SNIP on backend, you can ensure that there is no steering overhead for server side communication. While a vserver can be scaled on front-end as a Striped VIP, every node can have their own Spotted SNIPs to communicate to backed server farm ensuring least steering overhead on backend.
You see every possible use case is thought through while designing this beautiful architecture and we ensure that every deployment will have a right fit with NetScaler Cluster.