Hi all,

	I just thought I'd pop in my two cents on this whole topic.  In short 
form, on my Alpha GS80 (8 CPUs, 8 GB, 40 U160 disks/4x2channel adapters, 
2 NUMA nodes) 2.4.9 is almost unusable while 2.4.10 flies quite nicely. 
  Under 2.4.9 even things as simple as a mke2fs on a single 18 GB drive 
took minutes, while under 2.4.10 only seconds.  When I say 2.4.10, it's 
actually 2.4.10+the vm_tweak patch.

	I'm attaching a copy of the results of an AIM VII "shared" run as well as 
a lockstat report from a 500 user datapoint under 2.4.10.  If there is 
desire, I can generate a lockstat report for 2.4.9 as well.  I'm also 
going to see if I can get our profiling package to work on this system 
again, and if so I'll forward along those results as well.

	And not to bring up another old string, but just for giggles I removed 
the lock_kernel()/unlock_kernel() in llseek() to see what came of it. 
In short, a 2.2% gain in throughput and nearly 50% drop in the amount of 
time the kernel_lock is taken under this load.  Definitely looking 
forward to 2.5. ;)  Anyway, if there is something else anyone would be 
interested in that would be useful to run under this load (or a 
different set of statistics), feel free to let me know.

  - Pete