Thanks to cloud and the very generic "sticky sessions", many more people are aware of persistence as it relates to load balancing. It's a critical capability of load balancing without which stateful applications (which is most of them including VDI, most web applications, and data analysis tools) would simply fail to scale.
Persistence is, in general, like the many moods of Spock. They all look pretty much the same from the outside - ensure that a user, once connected, continues to be connected to the same application instance to ensure access to whatever state is stored in that instance.
But though they act the same (and Spock's expression appears the same) deep down, where it counts, persistence is very different depending on how it's implemented. It requires different processing, different inspection, different data, even. Understanding these differences is important because each one has a different impact on performance.
The Many Faces of Persistence
There are several industry de facto standard types of persistence: simple, SSL, and cookie. Then there are more advanced forms of persistence: SIP, WTS, Universal and Hash.
Generally speaking the de facto standard types of persistence are applicable for use with just about any web application. The more advanced forms of persistence are specific to a protocol or rely on a capability that is not necessarily standardized across load balancing services.
Without further adieu, let's dive in!
Simple persistence is generally based on network characteristics, like source IP address. It can also include the destination port, to give the load balancer a bit more capacity in terms of simultaneously applications supported. Best practices avoid simple persistence to avoid reoccurrence of the mega-proxy problem which had a tendency to overwhelm application instances. Network load balancing uses a form of simple persistence.
SSL Session ID Persistence
SSL Session ID persistence became necessary when SSL was broadly accepted as the de facto means of securing traffic in flight for web applications. Because SSL sessions need to be established and are very much tied to a session between client and server, failing to "stick" SSL-secured sessions results in renegotiation of the session, which takes a noticeable amount of time and annoys end-users. To avoid unnecessary renegotiation, load balancers use the SSL Session ID to ensure sessions are properly routed to the application instance to which they first connected.
Cookie persistence is a technique invented by F5 (shameless plug) that uses the HTTP cookie header to persist connections across a session. Most application servers insert a session id into responses that is used by developers to access data stored in the server session (shopping carts, etc... ). This value is used by load balancing services to enable persistence. This technique avoids the issues associated with simple persistence because the session id is unique.
Universal Persistence Universal persistence is the use of any piece of data (network, application protocol, payload) to persist a session. This technique requires the load balancer to be able to inspect and ultimately extract any piece of data from a request or response. This technique is the basis for application-specific persistence solutions addressing popular applications like SIP, WTS, and more recently, VMware View.
SIP, WTS, Username Persistence
Session Initiation Protocol (SIP) and Windows Terminal Server (WTS) persistence are application-specific persistence techniques that use data unique to a session to persist connections. Username persistence is a similar technique designed to address the needs of VDI - specifically VMware View solutions - in which sessions are persisted (as one might expect) based on username. When a type of persistence becomes very commonly used it is often moved from being a customized, universal persistence implementation to a native, productized persistence profile. This improves performance and scalability by removing the need to inspect and extract the values used to persist sessions from the data flow and results in an application-specific persistence type, such as SIP or WTS.
Hash persistence is the use of multiple values within a request to enable persistence. To avoid problems with simple persistence, for example, a hash value may be created based on Source IP, Destination IP, Destination Port. While not necessarily unique to every session, this technique results in a more even distribution of load across servers.
Non-unique value-based persistence techniques (simple, hash) are generally used with stateless applications or streaming content (video, audio) as a means to more evenly distribute load.
Unique value-based persistence techniques (universal, application-specific, SSL ID) are generally used with stateful applications that depend on the client being connected to the same application instance through the session's life.
Cookie persistence can be used with both techniques, provided the application is web based and uses HTTP headers for each request (Web Sockets breaks this technique).