openonload-20100923 Release Notes
=================================

 These release notes describe the major new features, and known limitations
 and bugs.


----------------------------------------
802.1Q VLANs
------------

 Traffic passed over a VLAN interface over a Solarflare port is
 accelerated.

 Traffic passed over a VLAN interface over a bonded interface is
 accelerated if the bond is accelerated.

 IMPORTANT: At time of writing we've just encountered a corner-case bug
            when using VLANs over bonds.  Debug is in progress.  We
            recommend you do not use VLANs over bonds in production in this
            release.

            We will make an update release in the near future.


----------------------------------------
Bonding / Teaming
-----------------

 Traffic passed over certain types of bonded interfaces is accelerated.

 What is accelerated?
  - Bonds containing only SFC interfaces.
  - Modes: 1 (active_backup), 2 (balance_xor), 4 (802.3ad).
  - xmit_hash_policy: layer2, layer2+3, layer3+4.

 Caveats: 
  - Bonding module v3.0.0 and later only.
  - Kernel must have sysfs support.

 Notes:
  - Bonds that are accelerated are accompanied by a message in the system
    log:
       [onload] Accelerating bond0 using OpenOnload

  - If a bond cannot be accelerated, there will be a message like this:
       [onload] Not accelerating bond0 using OpenOnload

  - An Onload application will not be able to use an interface that is
    added to a bond after the application has started.


----------------------------------------
Epoll
-----

 The epoll interfaces are now accelerated at user-level.  This is
 particularly valuable in applications that enable spinning (EF_POLL_USEC),
 since a user-level implementation is necessary to avoid interrupts and
 jitter, and minimise latency.

 poll() is still likely to give superior performance in many cases, because
 the epoll interface requires locking in order to protect its internal
 state, which imposes extra overhead.

 Known limitations:
  - Files are removed from the epoll set when the file descriptor is
    closed, rather than when the last reference to the file goes away.

  - Scalability when there are large numbers of sockets is relatively poor.

  - Deadlock is possible if a process fork()s concurrently with invoking
    epoll calls.

 Notes:
  - We are likely to do further work on epoll in future releases to address
    the limitations outlined above.

  - Acceleration can be disabled at runtime with EF_UL_EPOLL=0.  In that
    case epoll continues to work, but is handled in the kernel.