slow SMB speed

[b7842]

Config:
G4400
Asrock c236 wsi
32G ECC RAM
Intel® i210 onboard lan
WD 4TB*8 (raidz2)

Please kindly advise the following testing result.

dd if=/dev/zero of=/mnt/storage_vol/dataset_main/ddfile bs=2048k count=10000

CrytalMark to the zfspool

Ipref test result to the pfsense router

 

[UdoB]

Config:
G4400

Take a look on the CPUs on both sides while running CrystalDiskMark. Is there one core running with 100% load?

This is just a guess, but as far as I know Samba is single threaded per connection --> you can not transmit more data than one single CPU core can handle.

 

[b7842]

Take a look on the CPUs on both sides while running CrystalDiskMark. Is there one core running with 100% load?

This is just a guess, but as far as I know Samba is single threaded per connection --> you can not transmit more data than one single CPU core can handle.

I will test it tomorrow.
And my client pc config:
3770k
asus z77 deluxe
16GB ram

 

[joeschmuck]

What is the problem? Your Crystal Disk Mark looks very good, unless you are running 10Gb NICs of course.

EDIT: Sorry, I just re-read your posting, you were asking if this looks good. Yes it does look very good. You are maxing out your 1Gb Ethernet.

 

[DrKK]

Yes I agree with @joeschmuck, you are getting maximum performance out of a 1GbE.

 

[tvsjr]

You're maxing out gigabit Ethernet (minus overhead)... and your comparison to the dd results is likely flawed. Is that particular dataset running compression (lz4 being default)? If so, writing a bunch of zeros tells you nothing. 100GB of zeroes compresses quite well.

 

[b7842]

What is the problem? Your Crystal Disk Mark looks very good, unless you are running 10Gb NICs of course.

EDIT: Sorry, I just re-read your posting, you were asking if this looks good. Yes it does look very good. You are maxing out your 1Gb Ethernet.

For the c236 motherboard, there are two onboard lan.
They are i210 and i219.
On the intel spec sheet, the price of i210 is 3 times of i219.
I got 99 MB write speed per second by i219 in this morning.
I wonder to know that why the write speed cannot get > 110MB/s?
Because I got an omv system on the same network a few months ago and it gave me 10x write speed and 11x read speed.
OMV config:
4790k
32g ram
i350-t4
9211-8i hba with 4TB red*3
hosted on a esxi6 system

Anyway, I borrowed a i340-t2 lan card from my friend and i will test it later.

 

[b7842]

I don't know why the performance of i210 is slower than i219lm.

Later, i will place a i340-t2 into this machine and test it again.
tested @ Freenas9.10.2

 

[tvsjr]

If you're running gigabit, you're going to see 100MBps (pay attention to that capital B... megabytes), give or take. I suspect what you saw previously was 1,000Mbps (bits).

I suspect there are other things happening on both boxes at the same time, so you're not always going to see 100% of the theoretical maximum throughput. If you run Crystal for a few hours and average the results, I bet the 210 and 219 average out basically the same.

 

[joeschmuck]

When running Crystal Disk Mark, change the first number 1 (for how many tests it runs) to 5. Now run the test. A single test is not accurate at all.

As for the difference between the two interfaces, there is a hardware difference so there is going to be a minor difference.

Also, do you have compression turned on for your dataset? If yes, then create a new dataset without compression and try your testing again. If you are going to do benchmark tests then you need to ensure you are doing them properly. But in your specific situation I don't see any issue.

 

[b7842]

When running Crystal Disk Mark, change the first number 1 (for how many tests it runs) to 5. Now run the test. A single test is not accurate at all.

As for the difference between the two interfaces, there is a hardware difference so there is going to be a minor difference.

Also, do you have compression turned on for your dataset? If yes, then create a new dataset without compression and try your testing again. If you are going to do benchmark tests then you need to ensure you are doing them properly. But in your specific situation I don't see any issue.

finally the i350-t4 was delivered to me.
All hard wares were same, only the lan card changed to i350-t4.
That's why i don't accept the 80MB/s read performance.
For Giganetwork, it should get 9x% of the theoretical value because it is a quiet mature technology.

 

[Ericloewe]

Crystaldiskmark is not a network benchmark. Your earlier iperf test was just fine. The problem lies elsewhere.