Re: [RFC] nfs: use 4*rsize readahead size

[Bret Towe <magnade@gmail.com>]

On Mon, Mar 1, 2010 at 7:10 PM, Wu Fengguang <fengguang.wu@intel.com> wrote:
> Dave,
>
> Here is one more test on a big ext4 disk file:
>
>           16k  39.7 MB/s
>           32k  54.3 MB/s
>           64k  63.6 MB/s
>          128k  72.6 MB/s
>          256k  71.7 MB/s
> rsize ==> 512k  71.7 MB/s
>         1024k  72.2 MB/s
>         2048k  71.0 MB/s
>         4096k  73.0 MB/s
>         8192k  74.3 MB/s
>        16384k  74.5 MB/s
>
> It shows that >=128k client side readahead is enough for single disk
> case :) As for RAID configurations, I guess big server side readahead
> should be enough.
>
> #!/bin/sh
>
> file=/mnt/ext4_test/zero
> BDI=0:24
>
> for rasize in 16 32 64 128 256 512 1024 2048 4096 8192 16384
> do
>        echo $rasize > /sys/devices/virtual/bdi/$BDI/read_ahead_kb
>        echo readahead_size=${rasize}k
>        fadvise $file 0 0 dontneed
>        ssh p9 "fadvise $file 0 0 dontneed"
>        dd if=$file of=/dev/null bs=4k count=402400
> done

how do you determine which bdi to use? I skimmed thru
the filesystem in /sys and didn't see anything that says which is what

> Thanks,
> Fengguang
>
> On Fri, Feb 26, 2010 at 03:49:16PM +0800, Wu Fengguang wrote:
>> On Wed, Feb 24, 2010 at 03:39:40PM +0800, Dave Chinner wrote:
>> > On Wed, Feb 24, 2010 at 02:12:47PM +0800, Wu Fengguang wrote:
>> > > On Wed, Feb 24, 2010 at 01:22:15PM +0800, Dave Chinner wrote:
>> > > > What I'm trying to say is that while I agree with your premise that
>> > > > a 7.8MB readahead window is probably far larger than was ever
>> > > > intended, I disagree with your methodology and environment for
>> > > > selecting a better default value.  The default readahead value needs
>> > > > to work well in as many situations as possible, not just in perfect
>> > > > 1:1 client/server environment.
>> > >
>> > > Good points. It's imprudent to change a default value based on one
>> > > single benchmark. Need to collect more data, which may take time..
>> >
>> > Agreed - better to spend time now to get it right...
>>
>> I collected more data with large network latency as well as rsize=32k,
>> and updates the readahead size accordingly to 4*rsize.
>>
>> ===
>> nfs: use 2*rsize readahead size
>>
>> With default rsize=512k and NFS_MAX_READAHEAD=15, the current NFS
>> readahead size 512k*15=7680k is too large than necessary for typical
>> clients.
>>
>> On a e1000e--e1000e connection, I got the following numbers
>> (this reads sparse file from server and involves no disk IO)
>>
>> readahead size        normal          1ms+1ms         5ms+5ms         10ms+10ms(*)
>>          16k  35.5 MB/s        4.8 MB/s        2.1 MB/s       1.2 MB/s
>>          32k  54.3 MB/s        6.7 MB/s        3.6 MB/s       2.3 MB/s
>>          64k  64.1 MB/s       12.6 MB/s        6.5 MB/s       4.7 MB/s
>>         128k  70.5 MB/s       20.1 MB/s       11.9 MB/s       8.7 MB/s
>>         256k  74.6 MB/s       38.6 MB/s       21.3 MB/s      15.0 MB/s
>> rsize ==> 512k        77.4 MB/s       59.4 MB/s       39.8 MB/s      25.5 MB/s
>>        1024k  85.5 MB/s       77.9 MB/s       65.7 MB/s      43.0 MB/s
>>        2048k  86.8 MB/s       81.5 MB/s       84.1 MB/s      59.7 MB/s
>>        4096k  87.9 MB/s       77.4 MB/s       56.2 MB/s      59.2 MB/s
>>        8192k  89.0 MB/s       81.2 MB/s       78.0 MB/s      41.2 MB/s
>>       16384k  87.7 MB/s       85.8 MB/s       62.0 MB/s      56.5 MB/s
>>
>> readahead size        normal          1ms+1ms         5ms+5ms         10ms+10ms(*)
>>          16k  37.2 MB/s        6.4 MB/s        2.1 MB/s        1.2 MB/s
>> rsize ==>  32k        56.6 MB/s        6.8 MB/s        3.6 MB/s        2.3 MB/s
>>          64k  66.1 MB/s       12.7 MB/s        6.6 MB/s        4.7 MB/s
>>         128k  69.3 MB/s       22.0 MB/s       12.2 MB/s        8.9 MB/s
>>         256k  69.6 MB/s       41.8 MB/s       20.7 MB/s       14.7 MB/s
>>         512k  71.3 MB/s       54.1 MB/s       25.0 MB/s       16.9 MB/s
>> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>>        1024k  71.5 MB/s       48.4 MB/s       26.0 MB/s       16.7 MB/s
>>        2048k  71.7 MB/s       53.2 MB/s       25.3 MB/s       17.6 MB/s
>>        4096k  71.5 MB/s       50.4 MB/s       25.7 MB/s       17.1 MB/s
>>        8192k  71.1 MB/s       52.3 MB/s       26.3 MB/s       16.9 MB/s
>>       16384k  70.2 MB/s       56.6 MB/s       27.0 MB/s       16.8 MB/s
>>
>> (*) 10ms+10ms means to add delay on both client & server sides with
>>     # /sbin/tc qdisc change dev eth0 root netem delay 10ms
>>     The total >=20ms delay is so large for NFS, that a simple `vi some.sh`
>>     command takes a dozen seconds. Note that the actual delay reported
>>     by ping is larger, eg. for the 1ms+1ms case:
>>         rtt min/avg/max/mdev = 7.361/8.325/9.710/0.837 ms
>>
>>
>> So it seems that readahead_size=4*rsize (ie. keep 4 RPC requests in
>> flight) is able to get near full NFS bandwidth. Reducing the mulriple
>> from 15 to 4 not only makes the client side readahead size more sane
>> (2MB by default), but also reduces the disorderness of the server side
>> RPC read requests, which yeilds better server side readahead behavior.
>>
>> To avoid small readahead when the client mount with "-o rsize=32k" or
>> the server only supports rsize <= 32k, we take the max of 2*rsize and
>> default_backing_dev_info.ra_pages. The latter defaults to 512K, and can
>> be explicitly changed by user with kernel parameter "readahead=" and
>> runtime tunable "/sys/devices/virtual/bdi/default/read_ahead_kb" (which
>> takes effective for future NFS mounts).
>>
>> The test script is:
>>
>> #!/bin/sh
>>
>> file=/mnt/sparse
>> BDI=0:15
>>
>> for rasize in 16 32 64 128 256 512 1024 2048 4096 8192 16384
>> do
>>       echo 3 > /proc/sys/vm/drop_caches
>>       echo $rasize > /sys/devices/virtual/bdi/$BDI/read_ahead_kb
>>       echo readahead_size=${rasize}k
>>       dd if=$file of=/dev/null bs=4k count=1024000
>> done
>>
>> CC: Dave Chinner <david@fromorbit.com>
>> CC: Trond Myklebust <Trond.Myklebust@netapp.com>
>> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
>> ---
>>  fs/nfs/client.c   |    4 +++-
>>  fs/nfs/internal.h |    8 --------
>>  2 files changed, 3 insertions(+), 9 deletions(-)
>>
>> --- linux.orig/fs/nfs/client.c        2010-02-26 10:10:46.000000000 +0800
>> +++ linux/fs/nfs/client.c     2010-02-26 11:07:22.000000000 +0800
>> @@ -889,7 +889,9 @@ static void nfs_server_set_fsinfo(struct
>>       server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
>>
>>       server->backing_dev_info.name = "nfs";
>> -     server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
>> +     server->backing_dev_info.ra_pages = max_t(unsigned long,
>> +                                           default_backing_dev_info.ra_pages,
>> +                                           4 * server->rpages);
>>       server->backing_dev_info.capabilities |= BDI_CAP_ACCT_UNSTABLE;
>>
>>       if (server->wsize > max_rpc_payload)
>> --- linux.orig/fs/nfs/internal.h      2010-02-26 10:10:46.000000000 +0800
>> +++ linux/fs/nfs/internal.h   2010-02-26 11:07:07.000000000 +0800
>> @@ -10,14 +10,6 @@
>>
>>  struct nfs_string;
>>
>> -/* Maximum number of readahead requests
>> - * FIXME: this should really be a sysctl so that users may tune it to suit
>> - *        their needs. People that do NFS over a slow network, might for
>> - *        instance want to reduce it to something closer to 1 for improved
>> - *        interactive response.
>> - */
>> -#define NFS_MAX_READAHEAD    (RPC_DEF_SLOT_TABLE - 1)
>> -
>>  /*
>>   * Determine if sessions are in use.
>>   */
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>