Am I in the right ballpark r/w?

[heraclius]

I am currently teaching myself the ins and outs of FreeNAS. What I have setup it not production yet.

It is mostly recycled hardware, so it is pretty wrong compared to what is a "good idea". Once I am comfortable with FreeNAS something with ECC memory is likely.

AMD Athlon X2 240 (dual core 2.8Ghz)
2x4GB 1600 DDR3 NON-ECC
MSI 890 GD 70 (7 onboard sata)
2x Realtek NIC's (awesome I know, Intel 4x on the way)
5x 1Tb WD Blues

My desktop:
AMD FX-8320
16GB DDR3
Samsung 840 EVO
WD Green 3TB
Realtek NIC

Network is Gigabit.


I first setup 4x drives in a RAIDZ2. CIFS share.

Transferring 8Gb of data from my desktop's SSD to the Freenas copied at a constant rate of ~56MB/s. (according to my desktop)

Copying from the WD Green on my desktop to the NAS also copied at the same rate.

I killed the RAID, added the fifth drive, rebuilt the same config but with 5 drives, and tried again. No change in transfer speed. Copying to the NAS was ~56MB/s.

Copying from the NAS to my desktop SSD was ~45MB/s.

Why would reading from the NAS be slower than writing?

Why would I see no change in speed with the extra drive? The performance monitor during transfers doesn't show the cpu getting pegged at all. Maybe the realtek NIC's?

 

[Bidule0hm]

"Why would I see no change in speed with the extra drive?" It's (very) probably the Realtek NIC that limits you to ~50 MB/s. Even with an Intel NIC you'll don't see a change between 4 and 5 drives because the network will be the bottleneck (at about 120 MB/s). But at least with an Intel NIC you'll not have weird problems and you should saturate your gigabit link (if the network and the desktop NIC allow it of course).

 

[heraclius]

I guess I'll know more if the transfer speeds go up with a better NIC, I was expecting the network to cap it around 90MB/s...........

 

[mav@]

Why would reading from the NAS be slower than writing?

Because writes in ZFS are usually sequential and in most cases delayed, same time reads are often more random and depend on efficient read-ahead, that is not always predict perfectly.

 

[cyberjock]

To make matters worse, you aren't using hardware that is more than the bare minimum (and in some aspects not even recommended). Yes, I understand you are doing this for "testing", but it's a hell of a waste to try and test on hardware that is the bare minimum and not really even supported/recommended.

The harsh reality, if you want to do a test that actually means something, you gotta do it on hardware that is recommended. If you have anything resembling a thought that performance matters *at all* you shouldn't have the minimum 8GB of RAM.

 

[heraclius]

Welp, my network transfer speed appear to be caused by my desktop/Ubuntu. I rebooted into Windows 7 and had a ~105MB/s copy speed to the NAS for a 4GB file.

I am currently pricing ECC RAM.......to that end, is ZFS RAIDZ2 + non-ECC ram more dangerous than NTFS RAID 6 + non-ECC ram?

Btw, I am currently reading through your guide Cyber, it is very well written and organized. Thank you for putting that together.

 

[cyberjock]

I'm surprised that Ubuntu was slower. I run Linux Mint and Windows 7 dual boot with 10Gb LAN and I can totally smoke Windows 7 on CIFS shares, even when I use a RAM disk to avoid bottlenecks from the local disk.

You are running a fscking Realtek NIC on your desktop though, and Realteks really only work "great" on Windows.