Friday September 25, 2009
How Does Sailfin (Sun GlassFish Communications Server) SIP Session Replication Module Select Replica Instances?
For scalable deployments of middleware with high availability, employing a session state persistence approach to persist session state to all instances in the cluster could be a sub-optimal solution. Replicating sessions to all instances in the cluster would result in significantly higher network traffic just for replicating state reducing bandwidth for growing application user requests. This approach of sharing sessions across all instances perhaps is suited for small clusters with limited number of concurrent requests.
One of the better approaches to use to secure scaling advantages is the approach of buddy replication. In this approach, each instance selects one (or more) other instance(s) in the cluster to replicate any and all of its sessions. This is a superior approach and in fact, works for fairly large deployments. There are factors to consider here, in terms of the overhead the replication subsystem will need to handle at the cost of performance particularly when large number of concurrent sessions are being processed and later expired. An overhead to consider is the need for instances to form ring-like replica partnerships based on a certain order in which buddies would be available and selected. When a buddy instance fails, there is the cost of re-adjusting and forming new buddy relationship with another surviving instance, and when the original buddy recovers, to re-adjust again to use this upcoming instance as a replica partner by one of the instances in the cluster. Think of this as a chain based ring with its links randomly being removed but with the consistent goal of retaining a connected chain ring with the overhead of relinking each time a link is removed or added or a new one added to the chain.
There is also cost to be considered (if such were the design approach), each time the cluster shape changes for dynamically changing/updating any cookie information pertaining to replica locations that could be sent back as part of the response headers to the LB - typically that cost should also be avoided through more efficient means.
In the case of GlassFish, we have fairly successfully used buddy replication with each instance having a single replica buddy. We use the approach of locating sessions in the cluster when a request is directed by the LB to any random instance when a failure of an instance that was processing requests occurs. This has worked well for reasonably large mission critical environments where the scalability and availability requirements are within the boundaries of this approach.
In Sailfin 2.0, scaling and reliability needs for telco applications is typically very high and we needed a scalable approach to ensure Sip Session Replication overhead sustained good performance with the added reliability and availability. We, therefore, used a consistent hashing algorithm to dynamically assign a replica instance for each new session. This we did by leveraging the consistent hashing mechanism that the Sailfin Converged Load Balancer (CLB) uses for proxying requests to a target instance using a BEKey. In the case of replication, the same logic of using a hashed key for the target instance assignment is taken a bit further.
For replica selection, for each new session, we pre-calculate the most likely target instance that the CLB would failover to, if the current target primary instance that would serve the session, were to fail in future. This gives us the instance to which, the current primary instance, should replicate to. This gave us significant benefits in that there were no client cookie updates required to include replica partner information dynamically. There was no readjustment of replica partnerships needed when a particular instance failed as the hashing algorithm would provide another instance to replicate to with just an API call. When the failed instance comes back into the cluster, the sessions that were owned by it in its prior incarnation that are unexpired, would migrate back to it to maintain a balanced set of sessions across the cluster. And the replica selection algorithm would assign the original failover instance for this primary, as the replication partner.
Since this is based on a hashed selection algorithm with predetermined failover target, replica selection is dynamic, and does not need the knowledge of a particular order of instances being ready in the cluster to point all sessions from another instance as a replication partner. And more importantly, as the failover occurs to the specifc instance where replica data is located, there is significantly less network overhead to locate any particular session in the cluster when a particular request within the session scope is sent to the CLB. This allows for more bandwidth being available for a larger number of user sessions to be served. This approach is thus superior to the buddy replication approach and helped us scale to higher throughput and sustain a larger number of long running sessions.
It must be emphasized here that system level, and application server level tuning, and sizing are essential to ensure sustained performance, scalability and reliability in addition to the improvements provided with the SSR replication scheme and other parts of the Sailfin v2 server (aka Sun GlassFish Communications Server 2.0) .
As always, we welcome your feedback and encourage you to try Sailfin and send us any inputs and questions you may have in this respect.
Sailfin Promoted Builds are available here : Sailfin Downloads
Posted at 09:32AM Sep 25, 2009 by Shreedhar Ganapathy in GlassFish | Comments[0]
Thursday September 17, 2009
Project Shoal releases Shoal 1.1 Final Release
Project Shoal is announcing the release of Shoal 1.1 FCS today. This is the third release after Shoal 1.0 and 1.0 Update Release.
During the long time we have been working on this release, the Shoal team has accomplished many new features, and fixed many bugs providing significantly higher stability and reliability with the clustering core. A majority of the features and bug fixes were driven by the intense requirements coming out of the Telco grade application server from Project Sailfin (Sun GlassFish Communications Server) and Project GlassFish (Sun GlassFish Enterprise Server).
Some of the highlights of this release are :
- New event notification signal for JoinedAndReady members - applications using Shoal can call an API to report to other members that at the application level, the member is ready to begin operations - this distinguishes the member joining the group from the actual readiness of the application to start operations.
- New event notification signal for Group Leader changes - anytime a group leader leaves or fails, another group leader is selected and this signal allows Shoal GMS clients to be notified of this change to take any remedial actions.
- New health states Ready and AliveAndReady in the HealthMonitor member lifecycle state transitions to make it more closer to the finite state machine
- Introduce capability to distinguish between group startup and individual instance startup in JOIN and JOINEDANDREADY notifications. Applications may benefit knowing whether only a single GMS member is starting vs all GMS members of the GROUP are starting at once.
- Support for notifying failure within a specific threshold time when the failure is due to a network issue or a hardware end point failure. Blocking TCP health check connections would timeout in (default) 10 secs allowing for quicker detection of failure compared to default system settings for TCP Retransmission timeouts.
- A new GMS Watchdog Failure Notification API to be used by external service management frameworks (i.e. like GlassFish NodeAgent) - This allows external monitoring processes to notify GMS when these external process detects a member process failure faster than GMS heartbeat based detection can assign failures. This feature allows for faster failure detection notifications than the heartbeat based unreachability algorithm can determine failure. The feature was driven by the fact that the GlassFish Node Agent would detect a member process loss faster than Shoal could detect through its heartbeats and restart the failed process, resulting in no failure notifications from Shoal GMS to its clients. Full write up at document. Glassfish Issue 8308
This version of Shoal is already integrated into GlassFish v2.1.1 and Sailfin v2 both of which are upcoming releases slated for October 2009 end.
Details of the release announcement are here :
https://shoal.dev.java.net/servlets/NewsItemView?newsItemID=7645
You can download Shoal 1.1 binaries and sources with Javadocs here : https://shoal.dev.java.net/downloadsindex.html
As always, we owe it to you all in the community and thank you for all the feedback and issues you have raised to help us improve.
Thanks very much and hope to see your feedback continuing !
Posted at 12:44PM Sep 17, 2009 by Shreedhar Ganapathy in Shoal | Comments[0]
Tuesday September 08, 2009
Sun GlassFish Communications Server (Sailfin) adds High Availability Feature
Project Sailfin is building version 2.0 of the JSR 289 compliant application server. The Sailfin 2.0 release also known as Sun GlassFish Communications Server 2.0 will have a notable new feature with the addition of Sip Session Replication component. Sailfin 2.0 will provide High Availability of Sip artifacts providing resilience and availability of conversational state to Telco deployments. Sailfin 2.0 is targeted for release around end of October/early November 2009.
High availability through Sip Session Replication component (aka SSR component) allows for replication of SIP artifacts such as SIP Application Sessions, SIP Sessions, Timers, and Dialog Fragments in addition to Converged Sessions. Combined with existing GlassFish replication support for HTTP sessions, deployments can now be highly available covering both SIP protocol-only applications and converged (SIP and HTTP ) applications. To support the large scale load that can typically expected with Telco applications, the HA team employed a dynamic replica selection algorithm for each SIP artifact based on a consistent hashing algorithm thus obviating the need for buddy based replication where one would need to react to cluster shape changes and re-partner with another instance when a failure occurs - an expensive operation during high load conditions. (see this blog entry for more details )
The SSR component along with all of Sailfin is undergoing intensive quality testing including 24x7 longevity, scalability, reliability and fault tolerance testing at this time and we are making progress every day.
Turning on SSR in Sailfin 2.0 builds is extremely easy similar to how it is with GlassFish. You only need to deploy your SIP (JSR 289 compliant) or Converged Application with the availability-enabled option checked in the Admin GUI console or use the --availabilityenabled=true switch with the asadmin deploy command when deploy your SIP Archive sar file.
Here's your call to action : Go ahead and download the latest promoted build of Sailfin v2, deploy your SAR archive with availablity-enabled set to true (SSR enabled) and provide us feedback.
Posted at 08:00AM Sep 08, 2009 by Shreedhar Ganapathy in GlassFish | Comments[2]
This is a personal weblog, I do not speak for my employer.
