Using Linux Raid Systems

Shaohua Li | 5 Jul 2012 11:20

[patch 2/3 v4]raid1: read balance chooses idlest disk for SSD

SSD hasn't spindle, distance between requests means nothing. And the original
distance based algorithm sometimes can cause severe performance issue for SSD
raid.

Considering two thread groups, one accesses file A, the other access file B.
The first group will access one disk and the second will access the other disk,
because requests are near from one group and far between groups. In this case,
read balance might keep one disk very busy but the other relative idle.  For
SSD, we should try best to distribute requests to as more disks as possible.
There isn't spindle move penality anyway.

With below patch, I can see more than 50% throughput improvement sometimes
depending on workloads.

The only exception is small requests can be merged to a big request which
typically can drive higher throughput for SSD too. Such small requests are
sequential reads. Unlike hard disk, sequential read which can't be merged (for
example direct IO, or read without readahead) can be ignored for SSD. Again
there is no spindle move penality. readahead dispatches small requests and such
requests can be merged.

Last patch can help detect sequential read well, at least if concurrent read
number isn't greater than raid disk number. In that case, distance based
algorithm doesn't work well too.

V2: For hard disk and SSD mixed raid, doesn't use distance based algorithm for
random IO too. This makes the algorithm generic for raid with SSD.

Signed-off-by: Shaohua Li <shli <at> fusionio.com>
---

(Continue reading)

headers

Roberto Spadim | 5 Jul 2012 15:04

Re: [patch 2/3 v4]raid1: read balance chooses idlest disk for SSD

nice, just another question...
 since this use mixed raid disks (different types) could we improve
the algorithm to diferent hard disk speed, for example a raid1 with
7200 and 15000 rpm?
the distance continue the same but it will include a 'speed' factor
speed=1/rpm
(distance*speed)
or something to select fastest disk in the array
i don´t want to use write-mostly since it can reduce my total number
of read disks, but with this we could use the fastest disk with more
frequency without lose array 'speed'

2012/7/5 Shaohua Li <shli <at> kernel.org>
>
> SSD hasn't spindle, distance between requests means nothing. And the
> original
> distance based algorithm sometimes can cause severe performance issue for
> SSD
> raid.
>
> Considering two thread groups, one accesses file A, the other access file
> B.
> The first group will access one disk and the second will access the other
> disk,
> because requests are near from one group and far between groups. In this
> case,
> read balance might keep one disk very busy but the other relative idle.
> For
> SSD, we should try best to distribute requests to as more disks as
> possible.

(Continue reading)

headers

Shaohua Li | 6 Jul 2012 02:27

Re: [patch 2/3 v4]raid1: read balance chooses idlest disk for SSD

2012/7/5 Roberto Spadim <roberto <at> spadim.com.br>:
> nice, just another question...
>  since this use mixed raid disks (different types) could we improve
> the algorithm to diferent hard disk speed, for example a raid1 with
> 7200 and 15000 rpm?
> the distance continue the same but it will include a 'speed' factor
> speed=1/rpm
> (distance*speed)
> or something to select fastest disk in the array
> i don´t want to use write-mostly since it can reduce my total number
> of read disks, but with this we could use the fastest disk with more
> frequency without lose array 'speed'

Legitimate optimization, however detecting hard disk speed is hard.
Or maybe you want to add an option to tell kernel the disk speed?

Thanks,
Shaohua
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo <at> vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

headers

Roberto Spadim | 6 Jul 2012 04:45

Re: [patch 2/3 v4]raid1: read balance chooses idlest disk for SSD

well, i´m thinking something like elevators, they have some files to
configure options at /sys/block/sda/xxxx, it´s a nice interface...
echo and cat to configure the elevator or sysctl...

check this example of md in a old linux kernel (i don´t remember
version maybe 2.6.18)

#cd /sys/block/md0/md
# ls -l
-rw-r--r-- 1 root root 4096 Jul  5 23:39 array_state
--w------- 1 root root 4096 Jul  5 23:39 bitmap_set_bits
-rw-r--r-- 1 root root 4096 Jul  5 23:39 chunk_size
-rw-r--r-- 1 root root 4096 Jul  5 23:39 component_size
-r--r--r-- 1 root root 4096 Jul  5 23:39 degraded
drwxr-xr-x 2 root root    0 Jul  5 23:39 dev-sda1 <<<<=====
drwxr-xr-x 2 root root    0 Jul  5 23:39 dev-sdb1 <<<<=====
-rw-r--r-- 1 root root 4096 Jul  5 23:39 layout
-rw-r--r-- 1 root root 4096 Jul  5 23:39 level
-rw-r--r-- 1 root root 4096 Jul  5 23:39 metadata_version
-r--r--r-- 1 root root 4096 Jul  5 23:39 mismatch_cnt
--w------- 1 root root 4096 Jul  5 23:39 new_dev
-rw-r--r-- 1 root root 4096 Jul  5 23:39 raid_disks
lrwxrwxrwx 1 root root    0 Jul  5 23:39 rd2 -> dev-sda1
lrwxrwxrwx 1 root root    0 Jul  5 23:39 rd3 -> dev-sdb1
-rw-r--r-- 1 root root 4096 Jul  5 23:39 reshape_position
-rw-r--r-- 1 root root 4096 Jul  5 23:39 resync_start
-rw-r--r-- 1 root root 4096 Jul  5 23:39 safe_mode_delay
-rw-r--r-- 1 root root 4096 Jul  5 23:39 suspend_hi
-rw-r--r-- 1 root root 4096 Jul  5 23:39 suspend_lo
-rw-r--r-- 1 root root 4096 Jul  5 23:39 sync_action

(Continue reading)