(Solved) Performance decreased for no unknown reasons

[Caffe Mocha]

At first I installed Freenas on a SSD then I did some DD for performace test, I was able to hit around 400MB/s and the graphs in the reporting sections shows above 100MB/s per disk.
When I got my transcend USB from home, I did a fresh installation but my performance somehow dropped to 50MB~60 MB in the graphs , and DD test shows 110678105 bytes/sec.
Between Switching the SSD to USB, I have the switched my Sas cable to other cables but upon write test results I switched both the cables and SSD back.

I don't know what I did or didn't do that caused the performance drop. Hopefully someone could suggest me on what could cause the problem. Thank you



Motherboard: Supermicro A1SRM-2558F
RAM : 16G
HD: WD red 3 TB x 6 in RaidZ 2
HBA: IBM M1015 Firmware 16V

 

[pirateghost]

The medium in which freenas is installed on has zero bearing on the performance of the pool itself.

 

[Caffe Mocha]

thank you very much for your input , I am actually aware of the medium (USB or SSD) has no relation with performance itself
Just didn't want to leave stones unturned.
I wonder would a fresh installation would help with my situation?
PS how do I check if I turned on the encryption by mistake?

 

[Caffe Mocha]

problem solved. as it turns out, I set the record size to 4k when creating data and that's what slows down the performance

 

[wreedps]

Wow that is good performance. Are you using Iscsi?

 

[SweetAndLow]

You should try with a block size of 128k for the dd command and also update to the latest lsi controller firmware which is v20.

 

[Caffe Mocha]

Wow that is good performance. Are you using Iscsi?

it's NFS. Now I am trying to use Iscsi to backup Esxi.
still a newbie in Freenas, any good advice?
I have read that it's better to use Iscsi since Nfs with Esxi doesn't go well due to sync writes, unless SSD SLOG is used or disable sync writes.

 

[Caffe Mocha]

You should try with a block size of 128k for the dd command and also update to the latest lsi controller firmware which is v20.

Yes, Will update the Firmware to 20. and will surely give your suggestion a try. Just curious, does 128k block size help with actual real life performances?

I think FreeNAS community is a good crowd to hang out with.

 

[SweetAndLow]

What do you need real life? 128k is usually the zfs default block size I think so your buffet and chunk written to disk will be the same which is faster.

 

[monovitae]

problem solved. as it turns out, I set the record size to 4k when creating data and that's what slows down the performance

Out of curiosity could you(or anyone else for that matter) run the following iozone test from one of your nfs clients? I'm having some possibly similar odd isues with 4k and I'm curious what your results are.

Working directory should be from your nfs folder

Code:

iozone -s 2048m -r 4k -i 0 -i 2

 

[Caffe Mocha]

I have no idea how to help u in anyway but I inserted the your command in shell, and the result is in the pic.
hope that helps.

 

[monovitae]

Caffe thank you for assisting. Looking at your screen shot it looks like you ran it on the actual freenas box

Code:

root@freenas /mnt/backup/nfs

? If so this is still useful information because it shows us a baseline. However, what I would really need if possible is for you to run the same command from a NFS connected client, against the FreeNAS NFS share. Like if you have an ubuntu/mint/*nix machine or something similar.

On debian/ubuntu based systems you will likely need to install iozone

Code:

 sudo apt-get install iozone

 

[Caffe Mocha]

Ya, the ran it on freenas , must have skipped the "client " part in my head.
I will run the test again on Monday when I in office.