Content-Type: text/plain;
        charset=us-ascii
Mime-Version: 1.0 (Mac OS X Mail 11.5 \(3445.9.1\))
Subject: Re: [PATCH] SUNRPC: Remove rpc_xprt::tsh_size
From:   Chuck Lever <chuck.lever@oracle.com>
In-Reply-To: <e69a8e3965258109772f1adaedebc02d0863f3d9.camel@hammerspace.com>
Date:   Thu, 10 Jan 2019 12:13:21 -0500
Cc:     Linux NFS Mailing List <linux-nfs@vger.kernel.org>
Content-Transfer-Encoding: quoted-printable
Message-Id: <27A8300E-DF7E-48F3-8106-3027D0831C9C@oracle.com>
References: <20190103182649.4148.19838.stgit@manet.1015granger.net>
 <0331de80b8161f8bf16a92de20049cafb0c228da.camel@hammerspace.com>
 <90B38E07-3241-4CCD-A4C8-AB78BADFB0CD@oracle.com>
 <791EE189-59E5-4D58-9CF6-6D2CFC6C1210@oracle.com>
 <076cce85045dbaab3ca40947b2599f96cff66b53.camel@hammerspace.com>
 <1353EAC5-5BEE-461E-A11E-31F00FC7B946@oracle.com>
 <CE340D31-4F14-4C6F-B7C8-958754394C08@oracle.com>
 <1d6779ff05f2d31c4eccd048acbb28563bc9b79b.camel@hammerspace.com>
 <EE4DED67-E4A7-4F82-A5D0-9137381A2B4A@oracle.com>
 <e69a8e3965258109772f1adaedebc02d0863f3d9.camel@hammerspace.com>
To:     Trond Myklebust <trondmy@hammerspace.com>
Sender: linux-nfs-owner@vger.kernel.org
Precedence: bulk


> On Jan 4, 2019, at 5:44 PM, Trond Myklebust <trondmy@hammerspace.com> =
wrote:
>=20
> On Fri, 2019-01-04 at 16:35 -0500, Chuck Lever wrote:
>>> On Jan 3, 2019, at 11:00 PM, Trond Myklebust <
>>> trondmy@hammerspace.com> wrote:
>>>=20
>>> On Thu, 2019-01-03 at 17:49 -0500, Chuck Lever wrote:
>>>>> On Jan 3, 2019, at 4:35 PM, Chuck Lever <chuck.lever@oracle.com
>>>>>>=20
>>>>> wrote:
>>>>>=20
>>>>>> On Jan 3, 2019, at 4:28 PM, Trond Myklebust <
>>>>>> trondmy@hammerspace.com> wrote:
>>>>>>=20
>>>>>> On Thu, 2019-01-03 at 16:07 -0500, Chuck Lever wrote:
>>>>>>>> On Jan 3, 2019, at 3:53 PM, Chuck Lever <
>>>>>>>> chuck.lever@oracle.com>
>>>>>>>> wrote:
>>>>>>>>=20
>>>>>>>>> On Jan 3, 2019, at 1:47 PM, Trond Myklebust <
>>>>>>>>> trondmy@hammerspace.com> wrote:
>>>>>>>>>=20
>>>>>>>>> On Thu, 2019-01-03 at 13:29 -0500, Chuck Lever wrote:
>>>>>>>>>> +	reclen =3D req->rq_snd_buf.len;
>>>>>>>>>> +	marker =3D cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
>>>>>>>>>> reclen);
>>>>>>>>>> +	return kernel_sendmsg(transport->sock, &msg,
>>>>>>>>>> &iov, 1,
>>>>>>>>>> iov.iov_len);
>>>>>>>>>=20
>>>>>>>>> So what does this do for performance? I'd expect that
>>>>>>>>> adding
>>>>>>>>> another
>>>>>>>>> dive into the socket layer will come with penalties.
>>>>>>>>=20
>>>>>>>> NFSv3 on TCP, sec=3Dsys, 56Gbs IBoIP, v4.20 + my v4.21
>>>>>>>> patches
>>>>>>>> fio, 8KB random, 70% read, 30% write, 16 threads,
>>>>>>>> iodepth=3D16
>>>>>>>>=20
>>>>>>>> Without this patch:
>>>>>>>>=20
>>>>>>>> read: IOPS=3D28.7k, BW=3D224MiB/s
>>>>>>>> (235MB/s)(11.2GiB/51092msec)
>>>>>>>> write: IOPS=3D12.3k, BW=3D96.3MiB/s
>>>>>>>> (101MB/s)(4918MiB/51092msec)
>>>>>>>>=20
>>>>>>>> With this patch:
>>>>>>>>=20
>>>>>>>> read: IOPS=3D28.6k, BW=3D224MiB/s
>>>>>>>> (235MB/s)(11.2GiB/51276msec)
>>>>>>>> write: IOPS=3D12.3k, BW=3D95.8MiB/s
>>>>>>>> (100MB/s)(4914MiB/51276msec)
>>>>>>>>=20
>>>>>>>> Seems like that's in the noise.
>>>>>>>=20
>>>>>>> Sigh. That's because it was the same kernel. Again, with
>>>>>>> feeling:
>>>>>>>=20
>>>>>>> 4.20.0-rc7-00048-g9274254:
>>>>>>> read: IOPS=3D28.6k, BW=3D224MiB/s (235MB/s)(11.2GiB/51276msec)
>>>>>>> write: IOPS=3D12.3k, BW=3D95.8MiB/s
>>>>>>> (100MB/s)(4914MiB/51276msec)
>>>>>>>=20
>>>>>>> 4.20.0-rc7-00049-ga4dea15:
>>>>>>> read: IOPS=3D27.2k, BW=3D212MiB/s (223MB/s)(11.2GiB/53979msec)
>>>>>>> write: IOPS=3D11.7k, BW=3D91.1MiB/s
>>>>>>> (95.5MB/s)(4917MiB/53979msec)
>>>>>>>=20
>>>>>>=20
>>>>>> So about a 5% reduction in performance?
>>>>>=20
>>>>> On this workload, yes.
>>>>>=20
>>>>> Could send the record marker in xs_send_kvec with the head[0]
>>>>> iovec.
>>>>> I'm going to try that next.
>>>>=20
>>>> That helps:
>>>>=20
>>>> Linux 4.20.0-rc7-00049-g664f679 #651 SMP Thu Jan 3 17:35:26 EST
>>>> 2019
>>>>=20
>>>>  read: IOPS=3D28.7k, BW=3D224MiB/s (235MB/s)(11.2GiB/51185msec)
>>>> write: IOPS=3D12.3k, BW=3D96.1MiB/s (101MB/s)(4919MiB/51185msec)
>>>>=20
>>>=20
>>> Interesting... Perhaps we might be able to eke out a few more
>>> percent
>>> performance on file writes by also converting xs_send_pagedata() to
>>> use
>>> a single sock_sendmsg() w/ iov_iter rather than looping through
>>> several
>>> calls to sendpage()?
>>=20
>> IMO...
>>=20
>> For small requests (say, smaller than 17 pages), packing the head,
>> pagevec,
>> and tail into an iov_iter and sending them all via a single
>> sock_sendmsg
>> call would likely be efficient.
>>=20
>> For larger requests, other overheads would dominate. And you'd have
>> to keep around an iter array that held 257 entries... You could pass
>> a
>> large pagevec to sock_sendmsg in smaller chunks.
>>=20
>> Are you thinking of converting xs_sendpages (or even xdr_bufs) to use
>> iov_iter directly?
>=20
> For now, I was thinking of just converting xs_sendpages to call
> xdr_alloc_bvec(), and then do the equivalent of what xs_read_bvec()
> does for receives today.
>=20
> The next step is to convert xdr_bufs to use bvecs natively instead of
> having to allocate them to shadow the array of pages. I believe =
someone
> was working on allowing a single bvec to take an array of pages
> (containing contiguous data), which would make that conversion almost
> trivial.
>=20
> The final step would be to do as you say, to pack the kvecs into the
> same call to sock_sendmsg() as the bvecs. We might imagine adding a =
new
> type of iov_iter that can iterate over an array of struct iov_iter in
> order to deal with this case?

The same approach might help svc_send_common, and could be an easier
first step.


--
Chuck Lever