2006-05-08 14:14:19

by Shankar Anand

[permalink] [raw]
Subject: nfsstat for NFSv4

Hi,

I have added some code for NFSv4 server statistics for individual NFSv4
operations. I will submitting that to the NFS community soon.

I have some questions:
1. I noticed in Allen's patch for nfsstat for NFSv4 statistics that he
has taken into account only 32 NFSv4 operations. Do you have any reasons
for leaving out the other operations (like PUTFH, SECINFO, SAVEFH)?

2. I understand that in NFSv3 server we write the statistics into
/proc/net/rpc/nfsd in the same order as the RPC procedure numbers. Using
the same logic, I have written the NFSv4 statistics to be logged into
/proc/net/rpc/nfsd in the same order in which the operations are listed
in the NFSv4 RFC. But the NFSv4 client statistics that are getting
printed in Allen's patch seem to be of a different order. Why was this
decided to be so?

Thanks,
Shankar


2006-05-08 22:25:31

by J. Bruce Fields

[permalink] [raw]
Subject: Re: nfsstat for NFSv4

On Mon, May 08, 2006 at 08:14:19AM -0600, Shankar Anand wrote:
> I have added some code for NFSv4 server statistics for individual NFSv4
> operations. I will submitting that to the NFS community soon.
>
> I have some questions:
> 1. I noticed in Allen's patch for nfsstat for NFSv4 statistics that he
> has taken into account only 32 NFSv4 operations. Do you have any reasons
> for leaving out the other operations (like PUTFH, SECINFO, SAVEFH)?

At a guess: PUTFH and SAVEFH aren't particularly interesting, and
SECINFO isn't implemented yet on the server (and isn't used yet on the
client).

--b.

2006-05-09 13:52:55

by Shankar Anand

[permalink] [raw]
Subject: Re: nfsstat for NFSv4

Hi,

> "J. Bruce Fields" <[email protected]> 5/9/2006 3:55:31 am >>>
>>On Mon, May 08, 2006 at 08:14:19AM -0600, Shankar Anand wrote:
>> I have added some code for NFSv4 server statistics for individual
NFSv4
>> operations. I will submitting that to the NFS community soon.
>
>> I have some questions:
>> 1. I noticed in Allen's patch for nfsstat for NFSv4 statistics that
he
>> has taken into account only 32 NFSv4 operations. Do you have any
reasons
>> for leaving out the other operations (like PUTFH, SECINFO, SAVEFH)?


> At a guess: PUTFH and SAVEFH aren't particularly interesting, and
> SECINFO isn't implemented yet on the server (and isn't used yet on
the
> client).

But the problem with this is that we skip operations in between,
which breaks the order of the operations given in the RFC i.e. the
linear order
in which the statistics are captured in the NFSv4 server.
Just want to know why this was done when capturing the count for an
extra
procedure does not require much extra effort, going by the existing
code and
actually causes a break in the list of operation numbers.

>> 2. I understand that in NFSv3 server we write the statistics into
>> /proc/net/rpc/nfsd in the same order as the RPC procedure numbers.
Using
>> the same logic, I have written the NFSv4 statistics to be logged
into
>> /proc/net/rpc/nfsd in the same order in which the operations are
listed
>> in the NFSv4 RFC. But the NFSv4 client statistics that are getting
>> printed in Allen's patch seem to be of a different order. Why was
this
>> decided to be so?

It would be great if someone can answer this too.

Regards,
Shankar

2006-05-10 11:19:21

by Shankar Anand

[permalink] [raw]
Subject: Re: nfsstat for NFSv4

*Sorry I am repeating my previous email as it was badly cluttered*

Hi,

> "J. Bruce Fields" <[email protected]> 5/9/2006 3:55:31 am >>>
>> On Mon, May 08, 2006 at 08:14:19 AM Shankar Anand wrote:
>> I have added some code for NFSv4 server statistics for
>> individual NFSv4 operations. I will submitting that to the
>> NFS community soon.

>> I have some questions:
>> 1. I noticed in Allen's patch for nfsstat for NFSv4 statistics
>> that he has taken into account only 32 NFSv4 operations.
>> Do you have any reasons for leaving out the other
>> operations (like PUTFH, SECINFO, SAVEFH)?

> At a guess: PUTFH and SAVEFH aren't particularly interesting,
> and SECINFO isn't implemented yet on the server
> (and isn't used yet on the client).

But the problem with this is that we skip operations in
between, which breaks the order of the operations given
in the RFC i.e. the linear order in which the statistics are
captured in the NFSv4 server.

Just want to know why this was done when capturing
the count for an extra procedure does not require much
extra effort, going by the existing code and actually
causes a break in the list of operation numbers.

>> 2. I understand that in NFSv3 server we write the statistics
>> into /proc/net/rpc/nfsd in the same order as the RPC
>> procedure numbers. Using the same logic, I have written
>> the NFSv4 statistics to be logged into /proc/net/rpc/nfsd
>> in the same order in which the operations are listed in
>> the NFSv4 RFC.
>> But the NFSv4 client statistics that are getting printed in
>> Allen's patch seem to be of a different order.
>> Why was this decided to be so?

It would be great if someone can answer this too.

Regards,
Shankar