linux-kernel - Re: Bad SSD performance with recent kernels

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Sun, 29 Jan 2012 15:52:06 +0000
From:	Pádraig Brady <P@...igBrady.com>
To:	Eric Dumazet <eric.dumazet@...il.com>
CC:	Wu Fengguang <wfg@...ux.intel.com>,
	Herbert Poetzl <herbert@...hfloor.at>,
	Andrew Morton <akpm@...ux-foundation.org>,
	LKML <linux-kernel@...r.kernel.org>,
	Jens Axboe <axboe@...nel.dk>, Tejun Heo <tj@...nel.org>
Subject: Re: Bad SSD performance with recent kernels

On 01/29/2012 01:13 PM, Eric Dumazet wrote:
> Le dimanche 29 janvier 2012 à 19:16 +0800, Wu Fengguang a écrit :
> 
> 
>> Note that as long as buffered read(2) is used, it makes almost no
>> difference (well, at least for now) to do "dd bs=128k" or "dd bs=2MB":
>> the 128kb readahead size will be used underneath to submit read IO.
>>
> 
> Hmm...
> 
> # echo 3 >/proc/sys/vm/drop_caches ;dd if=/dev/sda of=/dev/null bs=128k count=32768
> 32768+0 enregistrements lus
> 32768+0 enregistrements écrits
> 4294967296 octets (4,3 GB) copiés, 20,7718 s, 207 MB/s
> 
> 
> # echo 3 >/proc/sys/vm/drop_caches ;dd if=/dev/sda of=/dev/null bs=2M count=2048
> 2048+0 enregistrements lus
> 2048+0 enregistrements écrits
> 4294967296 octets (4,3 GB) copiés, 27,7824 s, 155 MB/s

Same here on 2.6.40.4-5.fc15.x86_64
Note the SSD is rated for 500MB/s but is on a SATA II port,
and so limited by that. So the 128k result below is
close to the limit on this system.

Hmm, I previously tested this SSD with kernel-2.6.38.6-26.rc1.fc15.src.rpm
and got 270MB/s. Testing now gives variable and lower results:

# echo 3 >/proc/sys/vm/drop_caches; hdparm -tT /dev/sdb
/dev/sdb:
 Timing cached reads:   8388 MB in  2.00 seconds = 4200.73 MB/sec
 Timing buffered disk reads: 550 MB in  3.00 seconds = 183.19 MB/sec
# echo 3 >/proc/sys/vm/drop_caches; hdparm -tT /dev/sdb
/dev/sdb:
 Timing cached reads:   8260 MB in  2.00 seconds = 4134.30 MB/sec
 Timing buffered disk reads: 680 MB in  3.00 seconds = 226.63 MB/sec
# echo 3 >/proc/sys/vm/drop_caches; hdparm -tT /dev/sdb
/dev/sdb:
 Timing cached reads:   8426 MB in  2.00 seconds = 4217.87 MB/sec
 Timing buffered disk reads: 588 MB in  3.00 seconds = 195.96 MB/sec

Anyway testing different block sizes with dd:

# echo 3 >/proc/sys/vm/drop_caches; timeout -sINT 5 dd if=/dev/sdb of=/dev/null bs=2M
966787072 bytes (967 MB) copied, 5.00525 s, 193 MB/s
# echo 3 >/proc/sys/vm/drop_caches; timeout -sINT 5 dd if=/dev/sdb of=/dev/null bs=128k
1246494720 bytes (1.2 GB) copied, 4.99563 s, 250 MB/s

On a probably unrelated note, I've always noticed dd getting slower,
independent of disks, when the buffer size increases beyond 2M.

for i in $(seq 0 15); do
  size=$((16*1024**3)) #ensure this is big enough
  bs=$((1024*2**$i))
  printf "%8s=" $bs
  dd bs=$bs if=/dev/zero of=/dev/null count=$(($size/$bs)) 2>&1 |
  sed -n 's/.* \([0-9.]* [GM]B\/s\)/\1/p'
done

    1024=1.4 GB/s
    2048=2.6 GB/s
    4096=4.5 GB/s
    8192=6.7 GB/s
   16384=8.8 GB/s
   32768=9.4 GB/s
   65536=10.8 GB/s
  131072=11.5 GB/s
  262144=11.5 GB/s
  524288=11.3 GB/s
 1048576=11.3 GB/s
 2097152=10.6 GB/s
 4194304=6.5 GB/s
 8388608=5.9 GB/s
16777216=6.6 GB/s
33554432=6.6 GB/s

cheers,
Pádraig.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/