The following are examples of seeksize.d.
seeksize.d records disk head seek size for each operation by process.
This allows up to identify processes that are causing "random" disk
access and those causing "sequential" disk access.
It is desirable for processes to be accesing the disks in large
sequential operations. By using seeksize.d and bitesize.d we can
identify this behaviour.
In this example we read through a large file by copying it to a
remote server. Most of the seek sizes are zero, indicating sequential
access - and we would expect good performance from the disks
under these conditions,
# ./seeksize.d
Tracing... Hit Ctrl-C to end.
^C
22349 scp /dl/sol-10-b63-x86-v1.iso mars:\0
value ------------- Distribution ------------- count
-1 | 0
0 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ 726
1 | 0
2 | 0
4 | 0
8 |@ 13
16 | 4
32 | 0
64 | 0
128 | 2
256 | 3
512 | 4
1024 | 4
2048 | 3
4096 | 0
8192 | 3
16384 | 0
32768 | 1
65536 | 0
In this example we run find. The disk operations are fairly scattered,
as illustrated below by the volume of non sequential reads,
# ./seeksize.d
Tracing... Hit Ctrl-C to end.
^C
22399 find /var/sadm/pkg/\0
value ------------- Distribution ------------- count
-1 | 0
0 |@@@@@@@@@@@@@ 1475
1 | 0
2 | 44
4 |@ 77
8 |@@@ 286
16 |@@ 191
32 |@ 154
64 |@@ 173
128 |@@ 179
256 |@@ 201
512 |@@ 186
1024 |@@ 236
2048 |@@ 201
4096 |@@ 274
8192 |@@ 243
16384 |@ 154
32768 |@ 113
65536 |@@ 182
131072 |@ 81
262144 | 0
I found the following interesting. This time I gzipp'd the large file.
While zipping, the process is reading from one location and writing
to another. One might expect that as the program toggles between
reading from one location and writing to another, that often the
distance would be the same (depending on where UFS puts the new file),
# ./seeksize.d
Tracing... Hit Ctrl-C to end.
^C
22368 gzip sol-10-b63-x86-v1.iso\0
value ------------- Distribution ------------- count
-1 | 0
0 |@@@@@@@@@@@@ 353
1 | 0
2 | 0
4 | 0
8 | 7
16 | 4
32 | 2
64 | 4
128 | 14
256 | 3
512 | 3
1024 | 5
2048 | 1
4096 | 0
8192 | 3
16384 | 1
32768 | 1
65536 | 1
131072 | 1
262144 |@@@@@@@@ 249
524288 | 1
1048576 | 2
2097152 | 1
4194304 | 2
8388608 |@@@@@@@@@@@@@@@@@@ 536
16777216 | 0
The following example compares the operation of "find" with "tar".
Both are reading from the same location, and we would expect that
both programs would generally need to do the same number of seeks
to navigate the direttory tree (depending on caching); and tar
causing extra operations as it reads the file contents as well,
# ./seeksize.d
Tracing... Hit Ctrl-C to end.
^C
PID CMD
22278 find /etc\0
value ------------- Distribution ------------- count
-1 | 0
0 |@@@@@@@@@@@@@@@@@@@@ 251
1 | 0
2 |@ 8
4 | 5
8 |@ 10
16 |@ 10
32 |@ 10
64 |@ 9
128 |@ 11
256 |@ 14
512 |@@ 20
1024 |@ 10
2048 | 6
4096 |@ 7
8192 |@ 10
16384 |@ 16
32768 |@@ 21
65536 |@@ 28
131072 |@ 7
262144 |@ 14
524288 | 6
1048576 |@ 15
2097152 |@ 7
4194304 | 0
22282 tar cf /dev/null /etc\0
value ------------- Distribution ------------- count
-1 | 0
0 |@@@@@@@@@@ 397
1 | 0
2 | 8
4 | 14
8 | 16
16 |@ 24
32 |@ 29
64 |@@ 99
128 |@@ 73
256 |@@ 78
512 |@@@ 109
1024 |@@ 62
2048 |@@ 69
4096 |@@ 73
8192 |@@@ 113
16384 |@@ 81
32768 |@@@ 111
65536 |@@@ 108
131072 |@ 49
262144 |@ 33
524288 | 20
1048576 | 13
2097152 | 7
4194304 | 5
8388608 |@ 30
16777216 | 0