samba performance difference between old and the latest ?
Jeremy Allison
jra at samba.org
Fri Sep 6 16:18:45 UTC 2019
On Fri, Sep 06, 2019 at 11:01:51AM +0900, Namjae Jeon via samba-technical wrote:
> Hello,
>
> I found something strange during measuring performance with samba these days.
> I checked the performance of samba 4.7.6 and 4.10.6.
> There seems to be about 20% performance difference between the two versions.
>
> Test environment
> 1. use iozone and dd utils
> 2. share is tmpfs
> 3. smb client is kernel cifs client
> 4. oplock disable, oplock = no
> 4. command : ./iozone -s1g -r4 -e -i 0 -t 1 -b 24m.xls -+u
>
> with samba 4.7.6
> Children see throughput for 1 initial writers = 11248.48 kB/sec
> Parent sees throughput for 1 initial writers = 11248.28 kB/sec
> Min throughput per process = 11248.48 kB/sec
> Max throughput per process = 11248.48 kB/sec
> Avg throughput per process = 11248.48 kB/sec
> Min xfer = 1048576.00 kB
> CPU Utilization: Wall time 93.219 CPU time 29.517 CPU
> utilization
>
> with samba 4.10.6
> Children see throughput for 1 initial writers = 9018.82 kB/sec
> Parent sees throughput for 1 initial writers = 9018.70 kB/sec
> Min throughput per process = 9018.82 kB/sec
> Max throughput per process = 9018.82 kB/sec
> Avg throughput per process = 9018.82 kB/sec
> Min xfer = 1048576.00 kB
> CPU Utilization: Wall time 116.265 CPU time 33.182 CPU
> utilization
>
>
> ================================================================================
> ========
> dd test result is same with iozone result.
>
> samba 4.7.6 : 11.6MB/s
> samba 4.10.6 : 9.5MB/s
>
> with samba 4.7.6 :
> root at test1234-Samsung-DeskTop-System:/mnt/read# dd if=/dev/zero of=1.txt bs=4096
> count=1024
> 1024+0 records in
> 1024+0 records out
> 4194304 bytes (4.2 MB, 4.0 MiB) copied, 0.360991 s, 11.6 MB/s
>
> with samba 4.10.6 :
> root at test1234-Samsung-DeskTop-System:/mnt/read# dd if=/dev/zero of=1.txt bs=4096
> count=1024
> 1024+0 records in
> 1024+0 records out
> 4194304 bytes (4.2 MB, 4.0 MiB) copied, 0.442567 s, 9.5 MB/s
>
> Am I missing something? or real issue ?
Is this on identical kernel versions ?
More information about the samba-technical
mailing list