Stefan Richter wrote:
> Boaz Harrosh wrote:
>> From my tests, the block layer will, very fast, send you chained commands.
>> if you set it to 255 you'll see two chained fragments. In iscsi we have
>> 4096 and it is very good for performance. However since you are pre-allocating
>> the memory per can_queue commands, this can get big, and the memory overhead
>> can be an impact on performance.
>
> I did repeated read tests with dd now, with a 3.5" 750G SATA/ FireWire
> 800 disk and with a 2.5" 250G IDE/ FireWire 400 disk, each with either
> FireWire stack. I used dd with and without iflag=direct, just because I
> don't have a clue how to capture the influence of .sg_tablesize best.
>
> Tests with .sg_tablesize = 128, 255, and 5000 all yield the same
> transfer rates. So I guess I simply leave it at the current SG_ALL for now.
You will need to enable sg chaining for your device. I'm not sure
where is the latest documentation. (CC: Jens)
Here is the script I used in my tests:
#!/bin/bash
sdx=sdb
# mess with sglist chaining
cd /sys/block/$sdx/queue
echo 4096 > max_segments
cat max_hw_sectors_kb > max_sectors_kb
echo "max_hw_sectors_kb="$(cat max_hw_sectors_kb)
echo "max_sectors_kb="$(cat max_sectors_kb)
echo "max_segments="$(cat max_segments)
Try to view the content of max_segments before and after
see if it changed.
and also with dd you need a large block size here is how
I calculated it. (Other people might have better tests)
I did a write test but a read test should be better:
if=/dev/zero
of=/dev/sdb
outputfile=$1.txt
echo "Testing $1"
# send 32G in $1 sectrors at once
do_dd()
{
# blocks of one sector
bs=512
#memory page in blocks
page=8
#number of scatterlist elements in a transfer
sgs=$1
#calculate the bpt param
bpt=$(($sgs*$page))
#total blocks to transfer 32 Giga bytes
count=64M
echo $3: "bpt=$bpt"
\time bash -c \
"sg_dd blk_sgio=1 dio=1 if=$if of=$of bpt=$bpt bs=$bs count=$count 2>/dev/null" \
2>> $2
}
echo "BEGIN RUN $1" >> $outputfile
echo "run with 128-sgilist"
echo "run with 128-sgilist" >> $outputfile
for i in {1..20}; do
do_dd 128 $outputfile $i;
done
echo "run with 2048-sgilist"
echo "run with 2048-sgilist" >> $outputfile
for i in {1..20}; do
do_dd 2048 $outputfile $i;
done
I think that in the second case with 2048 if your device does
not support it, dd will fail. (if I recall correctly)
Boaz
PS: OK I used sg_dd from sg-utils sg_read might be even better
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
