Vrftjkry

I have a system with two 2TByte SATA disks configured as a Raid1 array.

There are times when the cpu is waiting for I/O for more than 20% of the time (output from sar) e.g.

09:25:01        CPU     %user     %nice   %system   %iowait    %steal     %idle

09:35:01        all     57,65      0,00      6,53     25,54      0,05     10,23

15:45:01        all      0,90      0,00      1,47     54,90      0,06     42,68

15:55:04        all      1,74      0,00      1,58     88,52      0,10      8,06

16:25:03        all      0,59      0,00      0,38     24,14      0,05     74,84

23:45:05        all      2,45      0,00      1,43     31,56      0,05     64,50

I collected additional information using atop which shows, that disk-I/O on one of the raid-disks is at the upper limit (disk sda, busy to 90%) e.g.:

MDD | md1 | busy 0% | | read 10174 | write 425 | | KiB/r 6 | KiB/w 7 | MBr/s 1.2 | | MBw/s 0.1 | avq 0.00 | | avio 0.00 ms |

DSK | sda | busy 90% | | read 9091 | write 507 | | KiB/r 6 | KiB/w 7 | MBr/s 0.9 | | MBw/s 0.1 | avq 1.14 | | avio 5.65 ms |

DSK | sdb | busy 18% | | read 1082 | write 507 | | KiB/r 11 | KiB/w 7 | MBr/s 0.2 | | MBw/s 0.1 | avq 1.39 | | avio 6.82 ms |

The man-page to atop states:

Such line shows the name (e.g. VolGroup00-lvtmp for a logical volume
or sda for a hard disk), the busy percentage i.e. the portion of time
that the unit was busy handling requests (busy), the number of read
requests issued (read), the number of write requests issued
(write), the number of KiBytes per read (KiB/r), the number of
KiBytes per write (KiB/w), the number of MiBytes per second
throughput for reads (MBr/s), the number of MiBytes per second
throughput for writes (MBw/s), the average queue depth (avq) and
the average number of milliseconds needed by a request (avio) for
seek, latency and data transfer.

Information can be read in parallel from both disks for raid1, but this is not used for a single stream of sequential input according to the md man page , explaining the fact that the second disk is not fully used

Looking at the MBr/s and MBw/s entries for sda, it looks like the disk is 90% busy with

0.9 + 0.1 MiBytes per second = 1 MiBytes per second = 8 MiBit per second

However, the expected rate for current disks is on the order of 1000 Mbit/s, which is roughly 100 times higher (neglecting the conversion from MiBit to Mbit).

The disks are (ouput of hdparm -I /dev/sda)

/dev/sda:



ATA device, with non-removable media

Model Number: TOSHIBA DT01ACA200

Serial Number: 54A8UH4GS

Firmware Revision: MX4OABB0

Transport: Serial, ATA8-AST, SATA 1.0a, SATA II Extensions, SATA Rev 2.5, SATA Rev 2.6, SATA Rev 3.0; Revision: ATA8-AST T13 Project D1697 Revision 0b

Standards:

Used: unknown (minor revision code 0x0029)

Supported: 8 7 6 5

Likely used: 8

Configuration:

Logical max current

cylinders 16383 16383

heads 16 16

sectors/track 63 63

--

CHS current addressable sectors: 16514064

LBA user addressable sectors: 268435455

LBA48 user addressable sectors: 3907029168

Logical Sector size: 512 bytes

Physical Sector size: 4096 bytes

Logical Sector-0 offset: 0 bytes

device size with M = 1024*1024: 1907729 MBytes

device size with M = 1000*1000: 2000398 MBytes (2000 GB)

cache/buffer size = unknown

Form Factor: 3.5 inch

Nominal Media Rotation Rate: 7200

Capabilities:

LBA, IORDY(can be disabled)

Queue depth: 32

Standby timer values: spec'd by Standard, no device specific minimum

R/W multiple sector transfer: Max = 16 Current = 16

Advanced power management level: disabled

DMA: mdma0 mdma1 mdma2 udma0 udma1 udma2 udma3 udma4 udma5 *udma6

Cycle time: min=120ns recommended=120ns

PIO: pio0 pio1 pio2 pio3 pio4

Cycle time: no flow control=120ns IORDY flow control=120ns

Commands/features:

Enabled Supported:

* SMART feature set

Security Mode feature set

* Power Management feature set

* Write cache

* Look-ahead

* Host Protected Area feature set

* WRITE_BUFFER command

* READ_BUFFER command

* NOP cmd

* DOWNLOAD_MICROCODE

Advanced Power Management feature set

Power-Up In Standby feature set

* SET_FEATURES required to spinup after power up

SET_MAX security extension

* 48-bit Address feature set

* Device Configuration Overlay feature set

* Mandatory FLUSH_CACHE

* FLUSH_CACHE_EXT

* SMART error logging

* SMART self-test

Media Card Pass-Through

* General Purpose Logging feature set

* WRITE_{DMA|MULTIPLE}_FUA_EXT

* 64-bit World wide name

* URG for READ_STREAM[_DMA]_EXT

* URG for WRITE_STREAM[_DMA]_EXT

* WRITE_UNCORRECTABLE_EXT command

* {READ,WRITE}_DMA_EXT_GPL commands

* Segmented DOWNLOAD_MICROCODE

* unknown 119[7]

* Gen1 signaling speed (1.5Gb/s)

* Gen2 signaling speed (3.0Gb/s)

* Gen3 signaling speed (6.0Gb/s)

* Native Command Queueing (NCQ)

* Host-initiated interface power management

* Phy event counters

* NCQ priority information

Non-Zero buffer offsets in DMA Setup FIS

* DMA Setup Auto-Activate optimization

Device-initiated interface power management

In-order data delivery

* Software settings preservation

* SMART Command Transport (SCT) feature set

* SCT Write Same (AC2)

* SCT Error Recovery Control (AC3)

* SCT Features Control (AC4)

* SCT Data Tables (AC5)

Security:

Master password revision code = 65534

supported

not enabled

not locked

frozen

not expired: security count

not supported: enhanced erase

320min for SECURITY ERASE UNIT.

Logical Unit WWN Device Identifier: 5000039ffac402a6

NAA : 5

IEEE OUI : 000039

Unique ID : ffac402a6

Checksum: correct

Is the output or the man-page for atop wrong or are the hard disks underperforming very much compared to the expected value or is there a missunderstanding on my side?

Or the wider question is: is my system really limited by the disk I/O capacity?

The expected sequential rate for current disks is on the order of 1000 Mbit/s, but this does not change anything for random IO.

A 7200 RPM disk will do around 120 random IOPS. So in a worst case scenario, where you always write only 1 byte, you will end up with a throughput of 120 Bytes/s.

Yes, this means, that there are around 3 decimal orders of magnitude between best case (only sequential) and worst case. Your result of 1KByte/s is somewhere in between, but much closer to the worst case.

You would profit very much from using an SSD: Even a small one when used as a bcache or LV cache can increase your random IOPS (especially those painfull small writes) by orders of magnitude.