On 7/26/17 5:47 AM, William Allen Simpson wrote:
On 7/14/17 1:08 PM, GerritHub wrote:
william.allen.simp...@gmail.com has uploaded this change for *review*.
View Change <https://review.gerrithub.io/369641>
Refresh call for review.
I'd already made this available for pre-review on Jul 10.
On 7/10/17 9:26 PM, William Allen Simpson wrote:
>
> This is only the dispatch side changes, mostly consisting of a pair
> of patches originally from 2015.
>
> The enqueue-dequeue code also used by 9P and upcalls is left intact.
> That should probably be the Next Big Thing,
>
There was a question that came up during the code walk-through yesterday.
Note that this patch removes calls to svc_rqst_rearm_events(). That is
now only called within nTI-RPC.
The old code only took new input after each request had been decoded,
resulting in a lot of data being backed up in the OS, and sawtooth
patterns in TCP timeouts and data retransmissions.
Now, events are handled more expeditiously. This means that we can
run 4 (or more) cores in parallel.
1) receiving data as soon as it arrives. This is carried as a
contiguous buffer or chain of buffers that can be zero-copy.
2) decoding the request input.
3) processing the request in the FSAL.
4) sending the request output. (Already in V2.5.)
Note also this is done without task switches, unless #3 has to wait for
the FSAL. One single thread runs seamlessly from receiving input to
sending output. Hopefully this improves processor cache consistency.
------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
Nfs-ganesha-devel mailing list
Nfs-ganesha-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs-ganesha-devel
