This provides virtual networking flexibility (add and remove when you want, have arbitrary number of network devices)
without having to compromise on performance, providing
dramatic improvements for applications that are dependent on network throughput
Major kudos to Raghuram Kothakota, the author of this valuable improvement. For further details, please see his blog article.
A new migration class has been defined that
supports these servers. Cross-CPU live migration required
"generic" CPU architecture definitions that disabled crypto acceleration; this is still used with the older T2, T2+ and T3 servers,
which cannot live migrate onto T5 and later. For further information, please see Eric Sharakan's blog article.
The SR-IOV and non-primary root domain features have firmware dependencies, which are documented in the Release Notes document.
Upgrading is really easy, especially for Solaris 11 users.
Just upgrade your control domain to the new SRU (Solaris 11.1 SRU 10.5), and it will automatically upgrade the logical domains daemon.
A patch will be available shortly to upgrade Solaris 10 systems.
This also replaces the virtual network device drivers with the version with improved performance.
The guest domains should also be upgraded to the new SRU and patch to gain the performance benefits described above, since
they are implemented in the Solaris drivers, and the command "ldm set-domain extended-mapin-space=on" needs to be set for service
and guest domains. That becomes a default for new domains, but may need to be set for domains you have already created.
That includes older T2 and T2+ lab servers that I manage with Oracle VM Manager. To upgrade a server, simply ran 'pkg update', and when ready to reboot into the new Solaris level. Before rebooting the server, I placed it in "maintenance mode" in the Oracle VM Manager user interface.
That caused guests to be migrated to another server in the same pool. After reboot, I took the server out of maintenance mode so the Manager
could freely use it for deploying guests. With very little effort I repeated this process for each lab server I work with. Even with such old equipment, I measured a close to 3x network performance improvement after the upgrade. Full performance will require a more recent T4 or later system.