* high cpu load for random write
@ 2009-06-30 12:54 Piavlo
2009-06-30 13:41 ` Chris Mason
0 siblings, 1 reply; 5+ messages in thread
From: Piavlo @ 2009-06-30 12:54 UTC (permalink / raw)
To: linux-btrfs
Hi,
I've just ran a tiobench benchmark on 2.6.31-rc1-git5 with
btrfs-progs-0.18
on single ATA disk with default mount options. Then performance now
looks great compared with previous kernel versions.
But one thing that I noticed - is that CPU load (besides being terribly
high) for random write - is several times higher than for sequential
write.
While for all other file systems ever tried, I've always seen the
opposite - the less data is written to the disk the less is the cpu load
(no matter if it's random or sequential writes).
I fail to understand why this is happening, could someone shed the
light on the issue?
Here is a sample output for one thread (this is a SMT enabled pentium4
single cpu system).
########################################
Tiotest results for 1 concurrent io threads:
,----------------------------------------------------------------------.
| Item | Time | Rate | Usr CPU | Sys CPU |
+-----------------------+----------+--------------+----------+---------+
| Write 9000 MBs | 189.1 s | 47.599 MB/s | 0.8 % |
38.9 % |
| Random Write 59 MBs | 17.6 s | 3.327 MB/s | 0.2 % | 89.1 % |
| Read 9000 MBs | 189.1 s | 47.589 MB/s | 0.6 % |
11.8 % |
| Random Read 59 MBs | 124.4 s | 0.471 MB/s | 0.0 % | 1.1 % |
`----------------------------------------------------------------------'
Tiotest latency results:
,-------------------------------------------------------------------------.
| Item | Average latency | Maximum latency | % >2 sec | % >10 sec |
+--------------+-----------------+-----------------+----------+-----------+
| Write | 0.077 ms | 2569.866 ms | 0.00009 | 0.00000 |
| Random Write | 0.022 ms | 0.970 ms | 0.00000 | 0.00000 |
| Read | 0.082 ms | 216.677 ms | 0.00000 | 0.00000 |
| Random Read | 8.293 ms | 124.126 ms | 0.00000 | 0.00000 |
|--------------+-----------------+-----------------+----------+-----------|
| Total | 0.106 ms | 2569.866 ms | 0.00004 | 0.00000 |
`--------------+-----------------+-----------------+----------+-----------'
#########################################
PS. Just for compare of ext4, brtfs & nilfs2 here are several runs results.
The script is:
------------------
#!/bin/bash
num_threads=$1
for x in /ext4 /btrfs /nilfs2; do
echo 3 > /proc/sys/vm/drop_caches
mount $x
rm -rf $x/*
echo "############################### $x
###############################"
tiotest -t $num_threads -f 9000 -r 15000 -b 4096 -d $x
umount $x
done
-----------------
The results for 1 io thread
------------------------------------------------------------
shell> ./test.tio 1
############################### /ext4 ###############################
Tiotest results for 1 concurrent io threads:
,----------------------------------------------------------------------.
| Item | Time | Rate | Usr CPU | Sys CPU |
+-----------------------+----------+--------------+----------+---------+
| Write 9000 MBs | 221.1 s | 40.707 MB/s | 0.8 % |
17.6 % |
| Random Write 59 MBs | 42.6 s | 1.374 MB/s | 0.0 % | 1.2 % |
| Read 9000 MBs | 215.5 s | 41.768 MB/s | 0.5 % |
7.7 % |
| Random Read 59 MBs | 123.2 s | 0.476 MB/s | 0.0 % | 0.8 % |
`----------------------------------------------------------------------'
Tiotest latency results:
,-------------------------------------------------------------------------.
| Item | Average latency | Maximum latency | % >2 sec | % >10 sec |
+--------------+-----------------+-----------------+----------+-----------+
| Write | 0.094 ms | 2812.766 ms | 0.00026 |
0.00000 |
| Random Write | 0.012 ms | 0.168 ms | 0.00000 |
0.00000 |
| Read | 0.093 ms | 164.746 ms | 0.00000
| 0.00000 |
| Random Read | 8.213 ms | 30.767 ms | 0.00000 | 0.00000 |
|--------------+-----------------+-----------------+----------+-----------|
| Total | 0.119 ms | 2812.766 ms | 0.00013 | 0.00000 |
`--------------+-----------------+-----------------+----------+-----------'
############################### /btrfs ###############################
Tiotest results for 1 concurrent io threads:
,----------------------------------------------------------------------.
| Item | Time | Rate | Usr CPU | Sys CPU |
+-----------------------+----------+--------------+----------+---------+
| Write 9000 MBs | 189.1 s | 47.599 MB/s | 0.8 % |
38.9 % |
| Random Write 59 MBs | 17.6 s | 3.327 MB/s | 0.2 % | 89.1 % |
| Read 9000 MBs | 189.1 s | 47.589 MB/s | 0.6 % |
11.8 % |
| Random Read 59 MBs | 124.4 s | 0.471 MB/s | 0.0 % | 1.1 % |
`----------------------------------------------------------------------'
Tiotest latency results:
,-------------------------------------------------------------------------.
| Item | Average latency | Maximum latency | % >2 sec | % >10 sec |
+--------------+-----------------+-----------------+----------+-----------+
| Write | 0.077 ms | 2569.866 ms | 0.00009 |
0.00000 |
| Random Write | 0.022 ms | 0.970 ms | 0.00000 |
0.00000 |
| Read | 0.082 ms | 216.677 ms | 0.00000
| 0.00000 |
| Random Read | 8.293 ms | 124.126 ms | 0.00000 | 0.00000 |
|--------------+-----------------+-----------------+----------+-----------|
| Total | 0.106 ms | 2569.866 ms | 0.00004 | 0.00000 |
`--------------+-----------------+-----------------+----------+-----------'
############################### /nilfs2 ###############################
Tiotest results for 1 concurrent io threads:
,----------------------------------------------------------------------.
| Item | Time | Rate | Usr CPU | Sys CPU |
+-----------------------+----------+--------------+----------+---------+
| Write 9000 MBs | 275.8 s | 32.628 MB/s | 0.5 % |
18.0 % |
| Random Write 59 MBs | 8.3 s | 7.087 MB/s | 0.0 % | 3.6 % |
| Read 9000 MBs | 251.5 s | 35.787 MB/s | 0.4 % |
7.1 % |
| Random Read 59 MBs | 129.8 s | 0.451 MB/s | 0.0 % | 0.8 % |
`----------------------------------------------------------------------'
Tiotest latency results:
,-------------------------------------------------------------------------.
| Item | Average latency | Maximum latency | % >2 sec | % >10 sec |
+--------------+-----------------+-----------------+----------+-----------+
| Write | 0.119 ms | 766.254 ms | 0.00000
| 0.00000 |
| Random Write | 0.501 ms | 1874.413 ms | 0.00000 | 0.00000 |
| Read | 0.109 ms | 267.389 ms | 0.00000
| 0.00000 |
| Random Read | 8.654 ms | 354.470 ms | 0.00000 | 0.00000 |
|--------------+-----------------+-----------------+----------+-----------|
| Total | 0.143 ms | 1874.413 ms | 0.00000 | 0.00000 |
`--------------+-----------------+-----------------+----------+-----------'
------------------------------------------------------------
and resuts for 2 io threads
------------------------------------------------------------
shell> ./test.tio 2
############################### /ext4 ###############################
Tiotest results for 2 concurrent io threads:
,----------------------------------------------------------------------.
| Item | Time | Rate | Usr CPU | Sys CPU |
+-----------------------+----------+--------------+----------+---------+
| Write 18000 MBs | 509.0 s | 35.365 MB/s | 1.5 % |
34.6 % |
| Random Write 117 MBs | 91.3 s | 1.284 MB/s | 0.0 % | 3.6 % |
| Read 18000 MBs | 483.5 s | 37.228 MB/s | 0.8 % |
13.2 % |
| Random Read 117 MBs | 266.6 s | 0.440 MB/s | 0.0 % | 1.6 % |
`----------------------------------------------------------------------'
Tiotest latency results:
,-------------------------------------------------------------------------.
| Item | Average latency | Maximum latency | % >2 sec | % >10 sec |
+--------------+-----------------+-----------------+----------+-----------+
| Write | 0.214 ms | 5004.055 ms | 0.00135 | 0.00000 |
| Random Write | 0.022 ms | 6.633 ms | 0.00000 | 0.00000 |
| Read | 0.209 ms | 444.293 ms | 0.00000 | 0.00000 |
| Random Read | 17.714 ms | 295.614 ms | 0.00000 | 0.00000 |
|--------------+-----------------+-----------------+----------+-----------|
| Total | 0.268 ms | 5004.055 ms | 0.00067 | 0.00000 |
`--------------+-----------------+-----------------+----------+-----------'
############################### /btrfs ###############################
Tiotest results for 2 concurrent io threads:
,----------------------------------------------------------------------.
| Item | Time | Rate | Usr CPU | Sys CPU |
+-----------------------+----------+--------------+----------+---------+
| Write 18000 MBs | 387.9 s | 46.405 MB/s | 1.9 % |
84.8 % |
| Random Write 117 MBs | 45.4 s | 2.583 MB/s | 0.1 % | 344.0 % |
| Read 18000 MBs | 438.7 s | 41.027 MB/s | 0.9 % |
19.3 % |
| Random Read 117 MBs | 268.2 s | 0.437 MB/s | 0.0 % | 2.1 % |
`----------------------------------------------------------------------'
Tiotest latency results:
,-------------------------------------------------------------------------.
| Item | Average latency | Maximum latency | % >2 sec | % >10 sec |
+--------------+-----------------+-----------------+----------+-----------+
| Write | 0.167 ms | 1607.292 ms | 0.00000 | 0.00000 |
| Random Write | 0.054 ms | 153.586 ms | 0.00000 | 0.00000 |
| Read | 0.190 ms | 562.645 ms | 0.00000 | 0.00000 |
| Random Read | 17.853 ms | 249.707 ms | 0.00000 | 0.00000 |
|--------------+-----------------+-----------------+----------+-----------|
| Total | 0.235 ms | 1607.292 ms | 0.00000 | 0.00000 |
`--------------+-----------------+-----------------+----------+-----------'
############################### /nilfs2 ###############################
Tiotest results for 2 concurrent io threads:
,----------------------------------------------------------------------.
| Item | Time | Rate | Usr CPU | Sys CPU |
+-----------------------+----------+--------------+----------+---------+
| Write 18000 MBs | 548.4 s | 32.825 MB/s | 1.4 % |
55.0 % |
| Random Write 117 MBs | 76.1 s | 1.539 MB/s | 0.1 % | 2.1 % |
| Read 18000 MBs | 1118.8 s | 16.088 MB/s | 0.4 % |
6.3 % |
| Random Read 117 MBs | 346.2 s | 0.338 MB/s | 0.0 % | 1.3 % |
`----------------------------------------------------------------------'
Tiotest latency results:
,-------------------------------------------------------------------------.
| Item | Average latency | Maximum latency | % >2 sec | % >10 sec |
+--------------+-----------------+-----------------+----------+-----------+
| Write | 0.237 ms | 1152.280 ms | 0.00000 | 0.00000 |
| Random Write | 4.966 ms | 1801.617 ms | 0.00000 | 0.00000 |
| Read | 0.485 ms | 502.967 ms | 0.00000 | 0.00000 |
| Random Read | 22.985 ms | 465.145 ms | 0.00000 | 0.00000 |
|--------------+-----------------+-----------------+----------+-----------|
| Total | 0.449 ms | 1801.617 ms | 0.00000 | 0.00000 |
`--------------+-----------------+-----------------+----------+-----------'
------------------------------------------------------------
^ permalink raw reply [flat|nested] 5+ messages in thread* Re: high cpu load for random write
2009-06-30 12:54 high cpu load for random write Piavlo
@ 2009-06-30 13:41 ` Chris Mason
2009-07-01 8:30 ` Piavlo
2009-07-01 8:51 ` Piavlo
0 siblings, 2 replies; 5+ messages in thread
From: Chris Mason @ 2009-06-30 13:41 UTC (permalink / raw)
To: Piavlo; +Cc: linux-btrfs
On Tue, Jun 30, 2009 at 03:54:09PM +0300, Piavlo wrote:
> Hi,
>
> I've just ran a tiobench benchmark on 2.6.31-rc1-git5 with
> btrfs-progs-0.18
> on single ATA disk with default mount options. Then performance now
> looks great compared with previous kernel versions.
>
> But one thing that I noticed - is that CPU load (besides being terribly
> high) for random write - is several times higher than for sequential
> write.
> While for all other file systems ever tried, I've always seen the
> opposite - the less data is written to the disk the less is the cpu load
> (no matter if it's random or sequential writes).
There are two causes of the high CPU load. The first is data
checksumming (which is constant for creating the file and for random
writes) and the second is the cost of maintaining back references for
the file data extent.
In btrfs, we track the owners of each extent, which makes repair, volume
management and other things much easier. Small random writes make for a
lot of extents, and so they also make for a lot of tracking.
In general, you'll find that mount -o ssd will be faster here, just
because it forces the allocator into more sequential allocations for
this workload.
You'll find that mount -o nodatacow uses much less CPU time, but this
disables checksumming and a few other advanced features.
-chris
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: high cpu load for random write
2009-06-30 13:41 ` Chris Mason
@ 2009-07-01 8:30 ` Piavlo
2009-07-01 15:53 ` Chris Mason
2009-07-01 8:51 ` Piavlo
1 sibling, 1 reply; 5+ messages in thread
From: Piavlo @ 2009-07-01 8:30 UTC (permalink / raw)
To: linux-btrfs
Hi Chris,
I've tried a test with mount -o nodatacow and see no noticeable
difference in cpu load.
Results with -o nodatacow (dmesg confirms "btrfs: setting nodatacow")
Tiotest results for 1 concurrent io threads:
,----------------------------------------------------------------------.
| Item | Time | Rate | Usr CPU | Sys CPU |
+-----------------------+----------+--------------+----------+---------+
| Write 9000 MBs | 195.7 s | 45.984 MB/s | 0.8 % | 39.2 % |
| Random Write 59 MBs | 23.7 s | 2.470 MB/s | 0.1 % | 91.6 % |
| Read 9000 MBs | 189.3 s | 47.538 MB/s | 0.6 % | 11.6 % |
| Random Read 59 MBs | 124.8 s | 0.470 MB/s | 0.0 % | 1.0 % |
`----------------------------------------------------------------------'
Tiotest latency results:
,-------------------------------------------------------------------------.
| Item | Average latency | Maximum latency | % >2 sec | % >10 sec |
+--------------+-----------------+-----------------+----------+-----------+
| Write | 0.079 ms | 3521.734 ms | 0.00009 | 0.00000 |
| Random Write | 0.021 ms | 0.449 ms | 0.00000 | 0.00000 |
| Read | 0.082 ms | 308.059 ms | 0.00000 | 0.00000 |
| Random Read | 8.318 ms | 31.541 ms | 0.00000 | 0.00000 |
|--------------+-----------------+-----------------+----------+-----------|
| Total | 0.107 ms | 3521.734 ms | 0.00004 | 0.00000 |
`--------------+-----------------+-----------------+----------+-----------'
Resutls without -o nodatacow
Tiotest results for 1 concurrent io threads:
,----------------------------------------------------------------------.
| Item | Time | Rate | Usr CPU | Sys CPU |
+-----------------------+----------+--------------+----------+---------+
| Write 9000 MBs | 199.4 s | 45.130 MB/s | 0.6 % | 39.1 % |
| Random Write 59 MBs | 23.8 s | 2.457 MB/s | 0.1 % | 92.1 % |
| Read 9000 MBs | 187.1 s | 48.110 MB/s | 0.5 % | 11.9 % |
| Random Read 59 MBs | 125.7 s | 0.466 MB/s | 0.0 % | 1.1 % |
`----------------------------------------------------------------------'
Tiotest latency results:
,-------------------------------------------------------------------------.
| Item | Average latency | Maximum latency | % >2 sec | % >10 sec |
+--------------+-----------------+-----------------+----------+-----------+
| Write | 0.081 ms | 3608.655 ms | 0.00026 | 0.00000 |
| Random Write | 0.021 ms | 1.032 ms | 0.00000 | 0.00000 |
| Read | 0.081 ms | 252.264 ms | 0.00000 | 0.00000 |
| Random Read | 8.376 ms | 192.319 ms | 0.00000 | 0.00000 |
|--------------+-----------------+-----------------+----------+-----------|
| Total | 0.107 ms | 3608.655 ms | 0.00013 | 0.00000 |
`--------------+-----------------+-----------------+----------+-----------'
Alex
Chris Mason wrote:
> There are two causes of the high CPU load. The first is data
> checksumming (which is constant for creating the file and for random
> writes) and the second is the cost of maintaining back references for
> the file data extent.
>
> In btrfs, we track the owners of each extent, which makes repair, volume
> management and other things much easier. Small random writes make for a
> lot of extents, and so they also make for a lot of tracking.
>
> In general, you'll find that mount -o ssd will be faster here, just
> because it forces the allocator into more sequential allocations for
> this workload.
>
> You'll find that mount -o nodatacow uses much less CPU time, but this
> disables checksumming and a few other advanced features.
>
> -chris
>
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: high cpu load for random write
2009-07-01 8:30 ` Piavlo
@ 2009-07-01 15:53 ` Chris Mason
0 siblings, 0 replies; 5+ messages in thread
From: Chris Mason @ 2009-07-01 15:53 UTC (permalink / raw)
To: Piavlo; +Cc: linux-btrfs
On Wed, Jul 01, 2009 at 11:30:23AM +0300, Piavlo wrote:
> Hi Chris,
>
> I've tried a test with mount -o nodatacow and see no noticeable
> difference in cpu load.
>
> Results with -o nodatacow (dmesg confirms "btrfs: setting nodatacow")
> Tiotest results for 1 concurrent io threads:
Well, it turns out that I broke nodatacow when I merged Christoph's attr
patch. This high CPU usage during tiotest is all checksums.
The CPU times look like random writes are using much more CPU, but
really they are just doing much less IO (59MB vs 9000MB). This makes
the CPU percentage look much higher because the test overall isn't
waiting on IO much at all.
Here are my results for your tiotest run with a fixed nodatacow:
test results for 1 concurrent io threads:
,----------------------------------------------------------------------.
| Item | Time | Rate | Usr CPU | Sys CPU |
+-----------------------+----------+--------------+----------+---------+
| Write 9000 MBs | 87.6 s | 102.774 MB/s | 0.4 % | 11.4 % |
| Random Write 59 MBs | 40.7 s | 1.438 MB/s | 0.0 % | 0.3 % |
`----------------------------------------------------------------------'
I'll queue the fix up for Linus.
-chris
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: high cpu load for random write
2009-06-30 13:41 ` Chris Mason
2009-07-01 8:30 ` Piavlo
@ 2009-07-01 8:51 ` Piavlo
1 sibling, 0 replies; 5+ messages in thread
From: Piavlo @ 2009-07-01 8:51 UTC (permalink / raw)
To: linux-btrfs
Chris Mason wrote:
> checksumming (which is constant for creating the file and for random
> writes) and the second is the cost of maintaining back references for
> the file data extent.
>
> In btrfs, we track the owners of each extent, which makes repair, volume
> management and other things much easier. Small random writes make for a
> lot of extents, and so they also make for a lot of tracking.
>
I've no problem with high cpu load on dedicated storage servers, but it
does not seems right for desktop usage.
Please correct me if i'm wrong.
Alex
> In general, you'll find that mount -o ssd will be faster here, just
> because it forces the allocator into more sequential allocations for
> this workload.
>
> You'll find that mount -o nodatacow uses much less CPU time, but this
> disables checksumming and a few other advanced features.
>
> -chris
>
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2009-07-01 15:53 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2009-06-30 12:54 high cpu load for random write Piavlo
2009-06-30 13:41 ` Chris Mason
2009-07-01 8:30 ` Piavlo
2009-07-01 15:53 ` Chris Mason
2009-07-01 8:51 ` Piavlo
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox