Re: NFS client latencies

[Ingo Molnar <mingo@elte.hu>]

* Andrew Morton <akpm@osdl.org> wrote:

> Well.  The radix-tree approach's best-case is probably quite a lot 
> worse than the list-based approach's best-case.  It hits a lot more 
> cachelines and involves a lot more code.

The list-based approach's best-case are large continuous append writes.  
No sorting overhead, and light data structures.

i'd say this workload should be not that bad under the radix tree either 
- the gang lookup stuffs a nice vector of 16 pages into an array.

we definitely can say nothing based on the observation that a _single_ 
page took 1.9 msecs in Lee's previous measurement, while 7700 pages now 
take 6 msecs to process.

> But of course the radix-tree's worst-case will be far better than 
> list's.

the generic VM/pagecache has proven that the radix tree wins hands down 
for alot more workloads than the worst-case.

> And presumably that list-based code rarely hits the worst-case, else 
> it would have been changed by now.

that was my other point in a previous mail: most write benchmarks do 
continuous append writes, and CPU overhead easily gets lost in network 
latency.

Also, considering that a good portion of the NFS client's code is still 
running under the BKL one would assume if the BKL hurts performance it 
would have been changed by now? ;-)

	Ingo