Message-ID: <4FBCE54F.5030205@panasas.com>
Date: Wed, 23 May 2012 16:25:35 +0300
From: Boaz Harrosh <bharrosh@panasas.com>
MIME-Version: 1.0
To: <andros@netapp.com>
CC: <trond.myklebust@netapp.com>, <linux-nfs@vger.kernel.org>
Subject: Re: [PATCH 4/4] NFSv4.1 test the mdsthreshold hint parameters
References: <1337763757-1566-1-git-send-email-andros@netapp.com> <1337763757-1566-4-git-send-email-andros@netapp.com>
In-Reply-To: <1337763757-1566-4-git-send-email-andros@netapp.com>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-nfs-owner@vger.kernel.org

On 05/23/2012 12:02 PM, andros@netapp.com wrote:

> From: Andy Adamson <andros@netapp.com>
> 
> Signed-off-by: Andy Adamson <andros@netapp.com>
> ---
>  fs/nfs/pnfs.c |   79 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++
>  1 files changed, 79 insertions(+), 0 deletions(-)
> 
> diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
> index 6620606..b8323aa 100644
> --- a/fs/nfs/pnfs.c
> +++ b/fs/nfs/pnfs.c
> @@ -936,6 +936,81 @@ pnfs_find_lseg(struct pnfs_layout_hdr *lo,
>  }
>  
>  /*
> + * Use mdsthreshold hints set at each OPEN to determine if I/O should go
> + * to the MDS or over pNFS
> + *
> + * The nfs_inode read_io and write_io fields are cumulative counters reset
> + * when there are no layout segments. Note that in pnfs_update_layout iomode
> + * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
> + * WRITE request.
> + *
> + * A return of true means use MDS I/O.
> + *
> + * From rfc 5661:
> + * If a file's size is smaller than the file size threshold, data accesses
> + * SHOULD be sent to the metadata server.  If an I/O request has a length that
> + * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
> + * server.  If both file size and I/O size are provided, the client SHOULD
> + * reach or exceed  both thresholds before sending its read or write
> + * requests to the data server.
> + */
> +static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
> +				     struct inode *ino, int iomode)
> +{
> +	struct nfs4_threshold *t = ctx->mdsthreshold;
> +	struct nfs_inode *nfsi = NFS_I(ino);
> +	loff_t fsize = i_size_read(ino);
> +	bool size = false, size_set = false, io = false, io_set = false, ret = false;
> +
> +	if (t == NULL)
> +		return ret;
> +
> +	dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
> +		__func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
> +
> +	switch (iomode) {
> +	case IOMODE_READ:
> +		if (t->bm & THRESHOLD_RD) {
> +			dprintk("%s fsize %llu\n", __func__, fsize);
> +			size_set = true;
> +			if (fsize < t->rd_sz)
> +				size = true;
> +		}
> +		if (t->bm & THRESHOLD_RD_IO) {
> +			dprintk("%s nfsi->read_io %llu\n", __func__,
> +				nfsi->read_io);
> +			io_set = true;
> +			if (nfsi->read_io < t->rd_io_sz)
> +				io = true;
> +		}
> +		break;
> +	case IOMODE_RW:
> +		if (t->bm & THRESHOLD_WR) {
> +			dprintk("%s fsize %llu\n", __func__, fsize);
> +			size_set = true;
> +			if (fsize < t->wr_sz)
> +				size = true;
> +		}
> +		if (t->bm & THRESHOLD_WR_IO) {
> +			dprintk("%s nfsi->write_io %llu\n", __func__,
> +				nfsi->write_io);
> +			io_set = true;
> +			if (nfsi->write_io < t->wr_io_sz)
> +				io = true;
> +		}
> +		break;
> +	}
> +	if (size_set && io_set) {
> +		if (size && io)
> +			ret = true;
> +	} else if (size || io)
> +		ret = true;
> +
> +	dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
> +	return ret;
> +}
> +
> +/*
>   * Layout segment is retreived from the server if not cached.
>   * The appropriate layout segment is referenced and returned to the caller.
>   */
> @@ -962,6 +1037,10 @@ pnfs_update_layout(struct inode *ino,
>  
>  	if (!pnfs_enabled_sb(NFS_SERVER(ino)))
>  		return NULL;
> +
> +	if (pnfs_within_mdsthreshold(ctx, ino, iomode))
> +		return NULL;
> +


Would we want to use these counters as the recommended layout_size in
read and write, instead of current's PAGE_SIZE?

Boaz

>  	spin_lock(&ino->i_lock);
>  	lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
>  	if (lo == NULL) {