FreeNAS not ready by a long shot

[progsup]

Here are the areas where FreeNAS really need improvement (and the reasons I am moving away from FreeNAS):

After the initial install the gigabyte network card operated at 10 Mb giving an 1.3 Mb throughput. Although I have skills and corrected the problem this was my first disappointment.

The second problem is power usage. Setting the hard drives to spin down resulted in AHCI timeout errors. There was no WebGUI setting to increase the timeout value. On top of that, the power daemon does seem to do anything and can not configured. Leaving me only with the "turn the system off by pressing the button" power saving option.

I expect FreeNAS out the box to go on standby after a configurable period of no network traffic over the shares. I expect it to tune-down the processor power when its not needed.

The third problem is CIFS/SMB performance out of the box. It should have easily congested my 100 Mb/s network bandwidth giving the hardware's disk and network performance which it did not. Although 80 MB/s write speed was decent, read speed did not get above the totally unacceptable 4,5 Mb/s.

On a positive note, I liked the WebGUI.

I ran the latest beta FreeNAS-8.0.1-RC2-amd64 (7813) on an ASUS P5QC Mobo, Dual Pentium CPU and 2 Gb of memory. Harddisks are six 1 TiB samsung F1 spinpoints configured as a single ZFSZ1.

 

[jgreco]

Here are the areas where FreeNAS really need improvement (and the reasons I am moving away from FreeNAS):

After the initial install the gigabyte network card operated at 10 Mb giving an 1.3 Mb throughput. Although I have skills and corrected the problem this was my first disappointment.

That sort of problem isn't all that unusual. FreeBSD does pretty well with automatic network detection, but quite frankly it's not trivial to get right for all cases. FreeNAS inherits most of that stuff from FreeBSD. Did you file a bug report?

The second problem is power usage. Setting the hard drives to spin down resulted in AHCI timeout errors. There was no WebGUI setting to increase the timeout value. On top of that, the power daemon does seem to do anything and can not configured. Leaving me only with the "turn the system off by pressing the button" power saving option.

The power daemon ratchets the clock speed up and down to make use of any power saving features that the CPU offers along those lines. It does not need "configuration" in order to work. It looks at the system load perhaps 10x a second and adjusts accordingly.

I expect FreeNAS out the box to go on standby after a configurable period of no network traffic over the shares.

That's not going to be a popular position; most of us want a NAS to have always-available storage.

I expect it to tune-down the processor power when its not needed.

It already does that if you ask it to, but you just finished saying that that doesn't do anything.

The third problem is CIFS/SMB performance out of the box. It should have easily congested my 100 Mb/s network bandwidth giving the hardware's disk and network performance which it did not. Although 80 MB/s write speed was decent, read speed did not get above the totally unacceptable 4,5 Mb/s.

On a positive note, I liked the WebGUI.

I ran the latest beta FreeNAS-8.0.1-RC2-amd64 (7813) on an ASUS P5QC Mobo, Dual Pentium CPU and 2 Gb of memory. Harddisks are six 1 TiB samsung F1 spinpoints configured as a single ZFSZ1.

So you didn't bother to read the hardware requirements and you don't like the performance you got for that failure.

What's an appropriate pithy comment here?

 

[louisk]

Not sure why you'd want the NAS to power off when its not being used. Hard to use it in that configuration. Not sure what your issue was with the network, I've never had an issue like that.
I agree that the write speed your seeing is too slow, although 2G of ram is not enough for ZFS (this is documented and falls short of the recommended minimum), and with RAIDZ, you will not get performance faster than a single spindle (this is also documented on how ZFS RAIDZ functions).

I think the biggest problem is that you had expectation that the system would do exactly what you wanted out of the box. I think this is unrealistic, regardless of the system. Any system will need adjustments to conform to your environment and preferences. I think that, for a free home system, there is nothing wrong with expecting the user to learn how the system works. Doesn't mean the user needs to be able to submit dev patches, but they shouldn't expect preconfigured. For a commercial environment, there is a paid version that comes with support and people who will work with you to get things working. This also seems very reasonable. If you buy an enterprise SAN, it doesn't come preconfigured for your environment out of the box, you have to do work to get there.

 

[progsup]

That sort of problem isn't all that unusual. FreeBSD does pretty well with automatic network detection, but quite frankly it's not trivial to get right for all cases. FreeNAS inherits most of that stuff from FreeBSD. Did you file a bug report?



The power daemon ratchets the clock speed up and down to make use of any power saving features that the CPU offers along those lines. It does not need "configuration" in order to work. It looks at the system load perhaps 10x a second and adjusts accordingly.



That's not going to be a popular position; most of us want a NAS to have always-available storage.



It already does that if you ask it to, but you just finished saying that that doesn't do anything.



So you didn't bother to read the hardware requirements and you don't like the performance you got for that failure.

What's an appropriate pithy comment here?

I am filing a bug report here, not intending to flame.

I know what the power daemon is supposed to do by default but it does not. I used a power consumption measuring device over 24 hrs and it never throttled down. Seems like a bug.

I did read the hardware requirements. Measuring the hardware's throughput using DD i got a nice 120 Mb/s. Writing to the server over my gigabyte Ethernet yielded a 80% use (80 Mb/s out of a possible 100 Mb/s). Reading from it however, yielded a meager 4,5 Mb/s. All this using CIFS/SMB. I am filing a bug about poor CIFS/SMB performance. Tweaking it trough the WebGUI did not help.

 

[progsup]

>>I agree that the write speed your seeing is too slow, although 2G of ram is not enough for ZFS (this is documented and falls short of the recommended minimum), and with RAIDZ, you will not get performance faster than a single spindle (this is also documented on how ZFS RAIDZ functions).

The write speed was fine. it was the read speed that is the problem.

>>I think the biggest problem is that you had expectation that the system would do exactly what you wanted out of the box.

I expected it to give me some grief. I expected to be able to resolve all issues through the webgui. I have listed the biggest caveats for me. Take your advantage with it.

 

[louisk]

What tuning options did you explore for CIFS/SMB? Did you upgrade the memory to the recommended minimum?

 

[progsup]

Hi Louisk,

Enabled RW and AIO. Disabled Sendfile. This saw me a 30 mb/s read speed through CIFS/SMB. But i have giving up. Currently I am running napp-it on OpenIndiana server edition (http://napp-it.org/doc/downloads/napp-it.pdf). The webGUI is not as good for dummies as FreeNAS is but reading the PDF helps. I can read at 55 Mb/s now using CIFS (and write at 85 Mb/s) on the same hardware without upgrades. Bonnie benchmark gives me 200 MB/s write and 498 MS/s read for sequential blocks. The gigabyte Ethernet with the CIFS protocol is the limiting factor here. OpenIndiana seems to be able to understand that my drives need a little time to spin-up. I still have to measure power consumption over an 24 hour period and test if it goes into S3 sleep.