While the graph is showing performance gains for native and virtual, its not necessarily meant to highlight differences between them. I've included both data sets so that people can see that large pages are not a virtualization-specific feature and provide benefits regardless of virtualization. The presence of support for LP unlocks value that the hardware, OS, and application already provide.
It is true, though, that there are slightly more gains to be had in a virtual environment than a physical environment. This is due to the fact that TLB misses cost a little bit more in a VM than in a native environment. It is possible that this trend reverses with the presence of a hardware MMU such as AMD RVI, however. But you'll have to read my (not-yet-published) article on AMD RVI for more details on that.
Scott,
Can you clarify this for me:
"Since applications such as Oracle databases and Java have been using large pages on Linux and Windows for years, the introduction of this support on ESX Server allows for increased gains in performance over previous virtual installs."
The graph then compares "Native" to "Virtual" - is this showing that Oracle and Java in a VM perform better than Oracle and Java on bare metal?
Thanks!
KLC
Ken Cline
Technical Director, Virtualization
Wells Landers
VMware Communities User Moderator