LinuxLists.cc - Re: Latest ext3 patches (extents, mballoc, delayed allocation)

2005-02-11 20:43:13

Subject: Re: Latest ext3 patches (extents, mballoc, delayed allocation)

Good day all,

I've updated the patchset against fixed and mballoc now behaves smarter patches collects some stats they not be used to store important
Please review. Any comments and
The patches are too big to send rejected). I put them in ftp://ftp.clusterfs.com/pub/people/alex/2.6.10
ack,reservation options
options
were used:
to 2 disks raid0
to 2 disks raid0
500M and 1000M using dd on fresh fs.
of disks, but sys.time is stable.
UP
real sys
19.03 2.50
18.99 4.19
15.08 4.16
19.07 2.52
21.23 3.09
19.25 2.67
44.59 5.02
40.77 8.31
40.51 8.30
38.74 4.93
44.88 6.08
39.98 5.27
/> UP
real sys
27.31 2.57
34.54 3.61
35.98 3.65
16.75 2.84
25.88 5.96
51.92 4.88
UP
real sys MB/s
5.93 1.54 63.79
3.60 1.85 131.372
3.55 1.85 133.071
2.28 1.53 150.403
2.88 2.11 119.815
2.87 1.91 119.564
9.96 6.22 137.316
11.49 7.40 130.146
11.45 7.38 130.467
9.25 6.31 147.872
11.85 8.59 115.392
11.98 7.69 114.177
18.32 12.14 149.228
23.77 14.88 119.995
23.35 15.04 123.088
21.00 12.49 130.223
24.98 17.67 114.083
25.84 15.53 105.743
47.37 25.13 115.582
53.43 30.78 102.266
54.68 30.44 100.082
44.89 25.73 121.962
56.48 36.41 96.9686
78.68 32.56 69.4764
108.1 51.24 101.354
139.84 63.77 78.8818
145.03 63.38 75.5155
100.24 52.50 109.26
128.77 77.94 85.7179
670.45 76.19 16.298
tiobench's results for SMP box:
Avg CPU
Rate (CPU%) Latency Eff
------ ------ --------- -----
1 40.95 12.80% 0.095 320
1 45.03 13.99% 0.086 322
1 50.53 15.18% 0.077 333
1 49.69 14.53% 0.078 342
1 29.56 12.08% 0.132 245
1 44.58 15.25% 0.087 292
2 22.36 13.21% 0.287 169
2 45.94 23.48% 0.108 196
2 46.29 23.57% 0.119 196
2 46.96 23.83% 0.111 197
2 23.05 14.89% 0.252 155
2 55.98 30.65% 0.064 183
4 45.39 55.10% 0.155 82
4 59.16 48.85% 0.103 121
4 54.35 46.83% 0.215 116
4 65.81 55.53% 0.119 119
4 26.35 32.38% 0.369 81
4 71.81 66.22% 0.113 108
Avg CPU
Rate (CPU%) Latency Eff
------ ------ --------- -----
1 119.05 40.37% 0.031 295
1 134.78 37.08% 0.028 363
1 25.18 8.377% 0.154 301
1 144.45 48.98% 0.026 295
1 33.53 13.51% 0.115 248
1 83.14 27.66% 0.045 301
2 34.28 19.08% -0.290 180
2 239.35 168.4% -0.549 142
2 257.29 167.9% -0.376 153
2 174.08 126.9% -1.885 137
2 45.32 25.80% -0.477 176
2 26.27 16.80% -0.762 156
4 138.70 173.9% -1.166 80
4 159.41 141.7% -1.310 112
4 89.80 87.91% -0.091 102
4 135.86 122.5% -3.192 111
4 109.81 153.8% -0.673 71
4 32.64 39.26% -1.949 83
Avg CPU
Rate (CPU%) Latency Eff
------ ------ --------- -----
1 55.15 37.13% 0.030 149
1 35.24 43.82% 0.106 80
1 34.24 42.86% 0.125 80
1 43.54 32.07% 0.028 136
1 51.67 84.37% 0.063 61
1 27.32 21.85% 0.057 125
2 18.76 34.45% 0.111 54
2 28.82 88.31% 0.236 33
2 30.63 91.98% 0.218 33
2 56.24 112.1% 0.018 50
2 18.31 85.35% 0.173 21
2 34.29 67.42% 0.213 51
4 15.92 54.67% 0.559 29
4 41.36 243.4% 0.137 17
4 45.44 275.5% 0.197 16
4 55.94 208.8% 0.020 27
4 20.47 176.9% 0.491 12
4 53.60 207.4% 0.056 26
Avg CPU
Rate (CPU%) Latency Eff
------ ------ --------- -----
1 0.73 0.565% 0.016 129
1 1.01 0.733% 0.018 137
1 1.01 0.742% 0.018 136
1 0.84 0.611% 0.015 137
1 0.74 0.531% 0.114 138
1 1.40 1.048% 0.018 134
2 1.21 2.248% 0.021 54
2 1.29 2.549% 0.021 51
2 1.20 2.448% 0.021 49
2 0.74 1.572% 0.018 47
2 1.28 2.325% 0.021 55
2 1.08 2.056% 0.020 52
4 1.04 3.760% 0.028 28
4 0.76 2.758% 0.029 28
4 0.80 3.052% 0.027 26
4 0.76 2.879% 0.024 26
4 1.13 3.854% 0.023 29
4 0.79 2.432% 0.025 32

2005-02-14 18:38:33

by Mingming Cao

[permalink] [raw]

Subject: Re: [Ext2-devel] Re: Latest ext3 patches (extents, mballoc, delayed allocation)

Alex Tomas wrote:

> Good day all,
>
> I've updated the patchset against 2.6.10. A bunch of bugs have been
> fixed and mballoc now behaves smarter a bit. Extents and mballoc
> patches collects some stats they print upon umount. NOTE: they must
> not be used to store important data. A lot of things are to be done.
>

Thanks Alex, for the hard work.

> Please review. Any comments and suggestions are very welcome.

Will do.

>
>
> The followins crazy listing shows tiobench's results for SMP box:
>
> Random Reads
> File Blk Num Avg CPU
> Identifier Size Size Thr Rate (CPU%) Latency Eff
> ---------------------------- ------ ----- --- ------ ------ --------- -----
> ext2 512 4096 1 119.05 40.37% 0.031 295
> ext3 512 4096 1 134.78 37.08% 0.028 363
> ext3rs 512 4096 1 25.18 8.377% 0.154 301

The throughput here is really weird. Reservation code does not touch
read code path. I could imagine that it maybe change the disk layout and
make a difference on sequential reads, but I am not sure how it will
affect the random read. And this is happening on 1 thread and 4 threads,
but for 2 threads, reservation case is the best. I will see if I could
repeat the same results here.

Mingming

2005-02-15 07:59:24

by Sonny Rao

[permalink] [raw]

Subject: Re: [Ext2-devel] Re: Latest ext3 patches (extents, mballoc, delayed allocation)

On Fri, Feb 11, 2005 at 11:40:21PM +0300, Alex Tomas wrote:
>
> Good day all,
>
> I've updated the patchset against 2.6.10. A bunch of bugs have been
> fixed and mballoc now behaves smarter a bit. Extents and mballoc
> patches collects some stats they print upon umount. NOTE: they must
> not be used to store important data. A lot of things are to be done.
>
> Please review. Any comments and suggestions are very welcome.

Alex, small buglet, If the FIBMAP-ioctl get's called on a file with
delayed allocation, you need to flush it (or at least allocate) before
returning the mappings. This doesn't seem to work properly at
present.

Sonny

2005-02-15 08:11:41

by Alex Tomas

[permalink] [raw]

Subject: Re: [Ext2-devel] Re: Latest ext3 patches (extents, mballoc, delayed allocation)

>>>>> Sonny Rao (SR) writes:

SR> Alex, small buglet, If the FIBMAP-ioctl get's called on a file with
SR> delayed allocation, you need to flush it (or at least allocate) before
SR> returning the mappings. This doesn't seem to work properly at
SR> present.

good catch. thanks.