SMB slow write fast read speed.

Hi quarros,

would you mind posting the HP ProLiant MicroServer Part Number so that we know what hardware we're talking about?

Is it this one?

https://www.cnet.com/products/…onitor-none-series/specs/

The first thing I would do is test the hdd using the simple dd command. It's not a perfect test but it will give you a good idea about the average sustained performance of each one of your drives.

1. login as root on your OMW console
2. find the path of your NFS share. using "df -Th" should list everything that is mounted, your NFS share should be in /export/nameofyourshare
3. cd in your NFS share
3. write the classic testfile by using "time dd if=/dev/zero of=testfile bs=4096 count=25000000" this will create a single 10GB file
4. now read the testfile "time dd if=testfile of=/dev/zero bs=4096 count=25000000"
5. convert the bytes in megabytes so that you have results that are easier to read (there are online converters that makes this quick)
6. you can connect via SSH and run a "top" command to see if something else but dd is using the ressources during write
7. repeat the process a few times (5) and of course do that for each drive if you have more than one

If dd is showing you an average of 80 MB/s writes and 120 MB/s you have a baseline to test your SAMBA share. You should have around 60-75 MB/s writes and 100-110 MB/s reads with some drops from time to time but you should be close to the numbers that dd has provided.

Use the testfile you created with dd above and copy it back and forth between your server and your clients as see what speed you have and what you can expect under the best circumstances/conditions.

Instead of wasting time with dd (and only testing 4K block sizes which might be heavily misleading since Windows 7 for example starts to write with 128K block sizes and up to 8 threads in parallel) I would better do this:

sudo apt install iozone3
cd $share
iozone -e -I -a -s 100M -r 4k -r 128k -r 1024k -i 0 -i 1

The most important parameter is the 2nd one which works on ext4 (but not on btrfs for example). Quoting the manual page:

-I     Use DIRECT IO if possible for all file operations. Tells the filesystem that all operations to the file are to bypass the buffer cache and go directly to disk. (not available on all platforms)

And the other test to perform is 'network isolated'. Use iperf3 between two Linux boxes or iperf/jperf.exe when a Windows client should be tested.