sg_rbuf.8

.TH SG_RBUF "8" "December 2001" "sg3_utils-0.96" SG_UTILS
.SH NAME
sg_rbuf \- reads data using SCSI READ BUFFER command
.SH SYNOPSIS
.B sg_rbuf
[[\fI-q\fR] | [\fI-d\fR] | [\fI-m\fR]] [\fI-t\fR] [\fI-s=<num_MB>\fR]
[\fI-b=<num_KB>\fR] \fI<sg_device>\fR
.SH DESCRIPTION
.\" Add any additional description here
.PP
This command reads data with the SCSI READ BUFFER command via
the sg driver and then discards it. Typically the data being 
read is from a disk's memory cache. It is assumed that the data
is sourced quickly (although this is not guaranteed by the SCSI
standards). This command is designed for timing transfer speeds
across a SCSI bus.
.TP
-q
only transfer the data into kernel buffers (typically by DMA from
the SCSI adapter card) and do not move it into the user space.
.TP
-d
use direct IO if available. In this case the sg driver will attempt
to configure the DMA from the SCSI adapter to transfer directly
into user memory. This will eliminate the copy via kernel buffers.
If not available then this will be reported and indirect IO will
be done instead.
.TP
-m
use memory mapped IO if available. In this case the sg driver will attempt
to configure the DMA from the SCSI adapter to transfer directly
into user memory. This will eliminate the copy via kernel buffers.
.TP
-t
times the bulk data transfer component of this command. The elapsed time
is printed out plus a MB/sec calculation. In this case "MB" is 1,000,000
bytes. The gettimeofday() system call is used internally for the time
calculation.
.TP
-s=<num_MB>
size of total transfer (in Megabytes == 1024^2). The default is
200 MB, the maximum is 4095 MB. The actual number of bytes
transferred may be slightly less since all transfers are the
same size (and an integer division is involved rounding towards zero).
.TP
-b=<num_KB>
size of each transfer (in Kilobytes == 1024). The default is
the actual available buffer size returned by the READ BUFFER
command. The maximum is the same as the default, hence this
argument can only be used to reduce the size of each transfer
to less than the device's actual available buffer size.
.PP
This command is typically used on modern SCSI disks which have
a RAM cache on their drive electronics. If no IO to the magnetic
media, or slower devices like flash RAM, is involved then
the disk should be able to source data fast enough to saturate
the bandwidth of the host machine. The bottleneck may then be
the SCSI bus, the Linux drivers or the host machine's hardware
(e.g. speed of RAM). Using time(1) in front of a
sg_rbuf command is one way to do such measurements.
.PP
Note that maximum bandwidths given by SCSI standards (e.g. UW ==
40 MB/sec) take "Mega" to mean 10^6 .
.SH EXAMPLES
.PP
On the test system /dev/sg0 corresponds to a fast disk
on a U2W SCSI bus (max 80 MB/sec). The disk specifications
state that its cache is 4 MB.
.br
   $ time ./sg_rbuf /dev/sg0
.br
READ BUFFER reports: buffer capacity=3434944,
.br
    offset boundary=6
.br
Read 200 MBytes (actual 199 MB, 209531584 bytes),
.br
    buffer size=3354 KBytes
.br
real 0m5.072s, user 0m0.000s, sys 0m2.280s
.PP
So that is approximately 40 MB/sec at 40 % utilization. Now with
the addition of the "-q" option this throughput improves and the
utilization drops to 0%.
.br
   $ time ./sg_rbuf -q /dev/sg0
.br
READ BUFFER reports: buffer capacity=3434944,
.br
    offset boundary=6
.br
Read 200 MBytes (actual 199 MB, 209531584 bytes),
.br
    buffer size=3354 KBytes
.br
real 0m2.784s, user 0m0.000s, sys 0m0.000s
.SH AUTHOR
Written by Doug Gilbert
.SH "REPORTING BUGS"
Report bugs to <dgilbert@interlog.com>.
.SH COPYRIGHT
Copyright \(co 2000, 2001 Douglas Gilbert
.br
This software is distributed under the GPL version 2. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
.SH "SEE ALSO"
.B time(1)