A user just showed up on IRC claiming this exact problem, but only on
CentOS/RHEL. I know it doesn't happen on Fedora. Maybe it's related
to default buffer sizes on sockets, or something?
On Thu, Oct 20, 2016 at 10:44 PM, Malahal Naineni <nain...@us.ibm.com> wrote:
> Please provide ganesha config and tcpdump that include FSINFO request (start
> tcpdump before the mount and do a very small I/O before you kill the
> tcpdump). Also, I am assuming that GPFS that deals with ganesha is not
> splitting the I/O as these traces don't indicate what ganesha is actually
> using, right?
>
> I know for a fact that we have seen I/Os with 1MB, so this can't be a
> ganesha hard limitation (including NFSv4 that I just experimented).
>
> Regards, Malahal.
>
>
> ----- Original message -----
> From: Malahal Naineni <mala...@gmail.com>
> To: Marc Eshel/Almaden/IBM@IBMUS
> Cc: Frank Filz <ffilz...@mindspring.com>, Matt Benjamin
> <mbenja...@redhat.com>, Malahal Naineni/Beaverton/IBM@IBMUS, NFS Ganesha
> Developers <nfs-ganesha-devel@lists.sourceforge.net>, Olaf Weiser
> <olaf.wei...@de.ibm.com>
> Subject: Re: [Nfs-ganesha-devel] nfs testing for SAP - status - new traces /
> new approach
> Date: Fri, Oct 21, 2016 7:55 AM
>
> I think that rpc code is about receive/send size which is quite
> different from NFS i/o size. Default max io size is 1M with GPFS fsal,
> but it is the client that could also limit. Since the same client is
> not limiting with kNFS, I am pretty sure you have something wrong in
> your ganesh config. The actual value should be provided to NFS client
> as part of FSINFO response. Please look at the FSINFO response with
> tcpdump.
>
> I know for a fact we have seen I/Os with 1MB at least with NFSv3. I
> will experiment with NFSv4 just in case, but I doubt this is due to
> code.
>
> On Fri, Oct 21, 2016 at 7:19 AM, Marc Eshel <es...@us.ibm.com> wrote:
>> We are not able to get IO bigger than 256K with Ganesha, same client and
>> kNFS can get 1M.
>> Is there something in Ganesha that limmits the IO size? see attached email
>> maxsize = 256 * 1024; /* XXX */, is that a problem ?
>> Marc.
>>
>>
>>
>> From: Sven Oehme/Almaden/IBM
>> To: Olaf Weiser/Germany/IBM@IBMDE
>> Cc: Malahal Naineni/Beaverton/IBM@IBMUS, dhil...@us.ibm.com,
>> fschm...@us.ibm.com, gfsch...@us.ibm.com, Marc Eshel/Almaden/IBM@IBMUS,
>> robg...@us.ibm.com
>> Date: 10/20/2016 05:32 PM
>> Subject: Re: nfs testing for SAP - status - new traces / new
>> approach
>>
>>
>> marc will send a email on what i found in the ganesha code. it seems that
>> max rpc size is hard limited to 256k :
>>
>> * Find the appropriate buffer size
>> */
>> u_int /*ARGSUSED*/
>> __rpc_get_t_size(int af, int proto, int size)
>> {
>> int maxsize, defsize;
>>
>> maxsize = 256 * 1024; /* XXX */
>> switch (proto) {
>> case IPPROTO_TCP:
>> defsize = 64 * 1024; /* XXX */
>> break;
>> case IPPROTO_UDP:
>> defsize = UDPMSGSIZE;
>> break;
>> default:
>> defsize = RPC_MAXDATASIZE;
>> break;
>> }
>> if (size == 0)
>> return defsize;
>>
>> /* Check whether the value is within the upper max limit */
>> return (size > maxsize ? (u_int) maxsize : (u_int) size);
>> }
>> 4:01:28 PM
>> in : src/rpc_generic.c
>>
>>
>> ------------------------------------------
>> Sven Oehme
>> Scalable Storage Research
>> email: oeh...@us.ibm.com
>> Phone: +1 (408) 824-8904
>> IBM Almaden Research Lab
>> ------------------------------------------
>>
>>
>>
>>
>>
>>
>>
>> ------------------------------------------------------------------------------
>> 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
>
>
>
>
>
> ------------------------------------------------------------------------------
> 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
>
------------------------------------------------------------------------------
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
