NFS Performance Tuning for HP-UX 11.0 and 11i Systems

nfs performance tuning for hp-ux 11.0 and 11i systems page 17
Notes:
Page 17July 22, 2002
Copyright 2002 Hewlett- Packard Company
local
filesystems
dd(1)
server(/big) -> timex dd if=/big/40gig of=/dev/zero bs=32k
1250000+0 records in
1250000+0 records out
real 5:36.58
user 4.90
sys 5:28.01
server(/big) -> timex dd if=/dev/zero of=/big/40gig bs=32k count=1250000
1250000+0 records in
1250000+0 records out
real 5:54.89
user 3.87
sys 5:50.65
Server Local File System Read
Performance = 121.7 Mb/sec
Server Local File System Write
Performance = 115.4 Mb/sec
Difference between 11.0 and 11i
“/dev/zero” is delivered with HP-UX 11i but not with HP-UX 11.0. It
can be created with the command “mknod /dev/zero c 3 4”
The dd(1) command was originally developed to facilitate the duplication of disk
drives by transferring their contents to magnetic tape. However, since it has the
ability to read from any input device and write to any output device, it is ideal for
generating disk I/O requests to specific filesystems. Also, since it allows the user to
control the record size used for the I/O transactions, dd can be used to simulate a
wide variety of applications.
When testing filesystem throughput, dd can be directed to read from or write to the
special file /dev/zero, thereby avoiding any physical disk read and write
requests. On 11.0 systems /dev/zero does not exist, but can be built via the
mknod(1M) command. Once built, reads from /dev/zero always returns a buffer
full of zeroes. The file is of infinite length, so the reading application needs to
specify the desired amount of data. Writes to /dev/zero are always successful,
but the written data is discarded.
NOTE: Avoid using the “ibs” and “obs” options to specify input and output block
sizes. When these options are specified, dd performs internal data conversions
that can dramatically impact the throughput numbers. Use the “bs” option instead.