inconsistent IOPS and throughput for /dev/mdX (?)

* inconsistent IOPS and throughput for /dev/mdX (?)
@ 2018-11-12  7:59 Paolo Valente
  0 siblings, 0 replies; only message in thread
From: Paolo Valente @ 2018-11-12  7:59 UTC (permalink / raw)
  To: Jens Axboe, linux-block, linux-kernel, linux-raid, Shaohua Li,
	Mike Snitzer, Coly Li, Michael Lyle, Mikulas Patocka,
	Heinz Mauelshagen, Tang Junhui
  Cc: Oleksandr Natalenko, Federico Motta, Ulf Hansson, Linus Walleij,
	Mark Brown

Hi,
the following command line

sudo fio --loops=20 --name=seqreader -rw=read --size=500M --numjobs=1 --filename=/dev/md0 & iostat -tmd /dev/md0 2 & sleep 15 ; killall iostat

generates a sequence of lines like this one:

------------------

8.11.2018 14:08:351)][15.4%][r=444MiB/s][r=114k IOPS][eta 00m:22s]

Device             tps    MB_read/s    MB_wrtn/s    MB_read    MB_wrtn
md0            1313,50       811,52         0,00       1623          0

------------------

So, iostat says that the throughput is about twice as high as that
actually enjoyed by fio.

The device layout is

$ lsblk
NAME                MAJ:MIN RM   SIZE RO TYPE   MOUNTPOINT
sda                   8:0    0 119,2G  0 disk
├─sda1                8:1    0   128M  0 part   /boot/EFI
└─sda2                8:2    0 119,1G  0 part
 └─md0               9:0    0 119,1G  0 raid10
   └─luks          253:0    0 119,1G  0 crypt
     ├─system-boot 253:1    0   512M  0 lvm    /boot
     ├─system-root 253:2    0 116,6G  0 lvm    /
     └─system-swap 253:3    0     2G  0 lvm    [SWAP]
sdb                   8:16   0 119,2G  0 disk
├─sdb1                8:17   0   128M  0 part
└─sdb2                8:18   0 119,1G  0 part
 └─md0               9:0    0 119,1G  0 raid10
   └─luks          253:0    0 119,1G  0 crypt
     ├─system-boot 253:1    0   512M  0 lvm    /boot
     ├─system-root 253:2    0 116,6G  0 lvm    /
     └─system-swap 253:3    0     2G  0 lvm    [SWAP]

The cause of the apparently inconsistent output of iostat are the numbers in /proc/diskstats:

  8      16 sdb 14175 991 2671357 19399 699083 167975 42916330 1241832 0251020 694150 840537 0 42682016 157759
  8      17 sdb1 57 0 4696 16 0 0 0 0 0 20 20 0 0 0 0
  8      18 sdb2 14078 991 2663981 19351 684072 167975 42916330 1128887 0240580 686070 840537 0 42682016 157759
  8       0 sda 14156 1225 2653954 21037 699566 167483 42916331 1151504 0202610 596520 840546 0 42682016 129468
  8       1 sda1 214 168 7776 80 1 0 1 0 0 60 60 0 0 0 0
  8       2 sda2 13902 1057 2643498 20935 684554 167483 42916330 1039618 0198390 595160 840546 0 42682016 129468
  9       0 md0 29891 0 8836730 0 740295 0 46361824 0 0 0 0 842371 0 46963528 0
  8      32 sdc 260 228 10536 1053 0 0 0 0 0 670 670 0 0 0 0
  8      33 sdc1 210 228 7856 797 0 0 0 0 0 490 490 0 0 0 0
  8      48 sdd 276 180 9624 1027 0 0 0 0 0 690 690 0 0 0 0
  8      49 sdd1 226 180 6944 763 0 0 0 0 0 480 480 0 0 0 0
  9     127 md127 654 0 9120 0 0 0 0 0 0 0 0 0 0 0 0
253       0 dm-0 24390 0 5300378 58830 733400 0 42791400 37540750 0 1196370 37938570 821551 0 42682016 338170
253       1 dm-1 183 0 6714 120 13 0 72 70 0 170 190 0 0 0 0
253       2 dm-2 24071 0 5287400 58960 725884 0 42791336 37486510 0 1205100 37892450 821551 0 42682016 346660
253       3 dm-3 94 0 4936 160 0 0 0 0 0 160 160 0 0 0 0

(In addition, here is /sys/block/md0/stat:
  29892        0  8836762        0   741031        0 46403088        0      0        0        0   843306        0 47004096        0)

Comparing the above numbers, we have

/dev/md0 : #read_io=29891 #sectors_read=8836730
/dev/sda2: #read_io=13902 #sectors_read=2643498
/dev/sdb2: #read_io=14078 #sectors_read=2663981

Is this a bug, or are we missing something?

Thanks,
Paolo

^ permalink raw reply	[flat|nested] only message in thread