Re: [PATCH v12 24/24] nfs: add FAQ section to Documentation/filesystems/nfs/localio.rst

[Jeff Layton <jlayton@kernel.org>]

On Mon, 2024-08-19 at 14:17 -0400, Mike Snitzer wrote:
> From: Trond Myklebust <trond.myklebust@hammerspace.com>
> 
> Add a FAQ section to give answers to questions that have been raised
> during review of the localio feature.
> 
> Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
> Co-developed-by: Mike Snitzer <snitzer@kernel.org>
> Signed-off-by: Mike Snitzer <snitzer@kernel.org>
> ---
>  Documentation/filesystems/nfs/localio.rst | 77 +++++++++++++++++++++++
>  1 file changed, 77 insertions(+)
> 
> diff --git a/Documentation/filesystems/nfs/localio.rst b/Documentation/filesystems/nfs/localio.rst
> index d8bdab88f1db..acd8f3e5d87a 100644
> --- a/Documentation/filesystems/nfs/localio.rst
> +++ b/Documentation/filesystems/nfs/localio.rst
> @@ -40,6 +40,83 @@ fio for 20 secs with 24 libaio threads, 128k directio reads, qd of 8,
>  - Without LOCALIO:
>    read: IOPS=12.0k, BW=1495MiB/s (1568MB/s)(29.2GiB/20015msec)
>  
> +FAQ
> +===
> +
> +1. What are the use cases for LOCALIO?
> +
> +   a. Workloads where the NFS client and server are on the same host
> +      realize improved IO performance. In particular, it is common when
> +      running containerised workloads for jobs to find themselves
> +      running on the same host as the knfsd server being used for
> +      storage.
> +
> +2. What are the requirements for LOCALIO?
> +
> +   a. Bypass use of the network RPC protocol as much as possible. This
> +      includes bypassing XDR and RPC for open, read, write and commit
> +      operations.
> +   b. Allow client and server to autonomously discover if they are
> +      running local to each other without making any assumptions about
> +      the local network topology.
> +   c. Support the use of containers by being compatible with relevant
> +      namespaces (e.g. network, user, mount).
> +   d. Support all versions of NFS. NFSv3 is of particular importance
> +      because it has wide enterprise usage and pNFS flexfiles makes use
> +      of it for the data path.
> +
> +3. Why doesn’t LOCALIO just compare IP addresses or hostnames when
> +   deciding if the NFS client and server are co-located on the same
> +   host?
> +
> +   Since one of the main use cases is containerised workloads, we cannot
> +   assume that IP addresses will be shared between the client and
> +   server. This sets up a requirement for a handshake protocol that
> +   needs to go over the same connection as the NFS traffic in order to
> +   identify that the client and the server really are running on the
> +   same host. The handshake uses a secret that is sent over the wire,
> +   and can be verified by both parties by comparing with a value stored
> +   in shared kernel memory if they are truly co-located.
> +
> +4. Does LOCALIO improve pNFS flexfiles?
> +
> +   Yes, LOCALIO complements pNFS flexfiles by allowing it to take
> +   advantage of NFS client and server locality.  Policy that initiates
> +   client IO as closely to the server where the data is stored naturally
> +   benefits from the data path optimization LOCALIO provides.
> +
> +5. Why not develop a new pNFS layout to enable LOCALIO?
> +
> +   A new pNFS layout could be developed, but doing so would put the
> +   onus on the server to somehow discover that the client is co-located
> +   when deciding to hand out the layout.
> +   There is value in a simpler approach (as provided by LOCALIO) that
> +   allows the NFS client to negotiate and leverage locality without
> +   requiring more elaborate modeling and discovery of such locality in a
> +   more centralized manner.
> +
> +6. Why is having the client perform a server-side file OPEN, without
> +   using RPC, beneficial?  Is the benefit pNFS specific?
> +
> +   Avoiding the use of XDR and RPC for file opens is beneficial to
> +   performance regardless of whether pNFS is used. However adding a
> +   requirement to go over the wire to do an open and/or close ends up
> +   negating any benefit of avoiding the wire for doing the I/O itself
> +   when we’re dealing with small files. There is no benefit to replacing
> +   the READ or WRITE with a new open and/or close operation that still
> +   needs to go over the wire.
> +
> +7. Why is LOCALIO only supported with UNIX Authentication (AUTH_UNIX)?
> +
> +   Strong authentication is usually tied to the connection itself. It
> +   works by establishing a context that is cached by the server, and
> +   that acts as the key for discovering the authorisation token, which
> +   can then be passed to rpc.mountd to complete the authentication
> +   process. On the other hand, in the case of AUTH_UNIX, the credential
> +   that was passed over the wire is used directly as the key in the
> +   upcall to rpc.mountd. This simplifies the authentication process, and
> +   so makes AUTH_UNIX easier to support.
> +
>  RPC
>  ===
>  

I'd just squash this into patch #19.

-- 
Jeff Layton <jlayton@kernel.org>