JGroups Protocol PING

External

• JGroups Manual http://www.jgroups.org/manual/html/protlist.html#PING
• JGroups Wiki http://community.jboss.org/wiki/JGroupsPING

Internal

• JGroups

TODO

• Next time I dive into this, come up with a full explanation for 'num_initial_members' below.

Overview

PING performs the initial (dirty) discovery of members, by multicasting PING request to an IP multicast address. The PING request is a Message with a PingHeader.GET_MBRS_REQ header. Each member replies with a (C, A) packet where C = coordinator address and A its own address. After a set number of milliseconds ('timeout') or replies ('num_initial_members'), the joiner determines the coordinator from responses, and sends a JOIN to it. The JOIN request is handled by the GMS layer. If nobody responds, the joiner assumes it's the first member of a group.

Configuration

  <PING timeout="2000" num_initial_members="3"/>

Difference between PING and MPING

MPING, unlike PING, uses its own IP multicast socket to send discovery requests. PING simply sends a discover event down the stack to be sent by the underlying transport protocol.

Periodic GET_MBRS_REQ Requests

The super-class Discovery protocol is in charge with sending GET_MBRS_REQ (get members) requests upon receiving a FIND_ALL_VIEWS , FIND_INITIAL_MBRS, etc. event. The "find" events are sent either by the GMS protocol, when initializing, or periodically by the FindSubgroupsTask of the MERGE2 protocol.

The discovery requests are generated by the PingSenderTask, in the form of OOB messages with PingHeader.GET_MBRS_REQ headers. PingSenderTask sends these messages in a delayed loop. The delay between requests is timeout /num_ping_requests where timeout and num_ping_requests are configuration parameters described below. Normally that should be about 1.5 sec (1,500 ms).

The loop is interrupted either after a 'timeout' milliseconds (which means that all 'num_ping_requests' discovery requests have been sent), or after the first valid response - which allows determining who the coordinator is - is received. Further clarifications needed as to how 'num_initial_members' works.

A new PingSenderTask - and the loop - will be started again when a new GET_MBRS_REQ message needs to be sent.

The result of the GET_MBRS_REQ process is a list of ping responses (see PingData).

Stack Boot Time Factor

The whole stack is waiting while the first discovery loop is running, setting the 'timeout' to a smaller value can reduce the stack boot time.

Configuration

timeout

The timeout to wait for the initial members. Default is 3000 ms.

num_initial_members

TODO

num_ping_requests

Number of discovery requests to be sent distributed over timeout. Default is 2.

Filtering Out

If you want to filter these messages out, the header id is 6.