Is it a good idea to run freenas 8 (inside virtualbox) on a lamp server?

Question

I am redesigning my home network setup (specifically my nas and server), and recently asked a question about running a lamp server and freenas 8 on two separate machines. However, in the spirit of conserving money and electricity, i am thinking of combining the two machines using virtualbox. This way, i could run freenas 8 in a virtualbox on my lamp server (on a single machine).

To prevent any conflicts between the two OS's I thought i could install both the lamp server and freenas 8 on a single usb-drive (connected to an internal usb header), and have a couple of hd's that only freenas accesses for storage (so there won't be any competition between linux and freenas for the hd's).

Is this potentially a setup that could work and remain stable? And is it sensible to run an OS from flash memory for long periods of time (because it only has a limited number of read/writes), or is it an accident waiting to happen? Any possible suggestions or foreseen issues?

Answer 1

Why don't you go for Xen directly ?
You'll get a much better use and sharing of ressources and if your fedora crashes you'll always have your NAS available.

Here are some HOWTOs on setting up a Xen server:
http://wiki.xensource.com/xenwiki/HowTos

...And you could put a nice pfSense instance to firewall all that traffic properly :)

Answer 2

I'm running freenas 8.0.4 as an HVM DomU using a xen 4.0 hypervisor and a debian squeeze Dom0 (with an amd cpu if that matters). I had no problems setting it up and everything seems to be running flawlessly. So it works and could be an option. So far, with limited testing, performance seems good -- (At least good enough given my current network and load).

Now, back to your question: "Is it a good idea?".

My answer would be if the performance is the same or close to running freenas native, it would be a good idea. I haven't had enough time (or enough memory) to conduct a fair comparison test. I'm interested in that answer myself.

I'll post when I run some more tests.

--------------------- added after more testing -----------------------------

OK. My answer is that it isn't a good idea. I used rsync for most of the tests since it reports "Bytes/sec" and that made it easy. I compared that figure to my graphical network monitor and the reporting graphs inside freenas. All the data agreed.

Sending files to/from the Dom0 Linux machine, I could easily saturate a 100Mb/s ethernet link. I had reported transfer speeds both to and from the host freenas was running on at just over 11 million Bytes/second.

Sending files to/from the DomU freenas machine, I only achived about half that. Speeds to/from averaged about 4.5 Million Bytes/second. CPU load on both the freenas vm and the Dom0 host were very high. I didn't even bother repeating the test with a gigabit ethernet link since the results seemed unambiguous.

With an HVM machine, my understanding is the machine should run exactly as fast as the underlying hardware on high cpu tasks. But the device drivers are emulated. In this case, the entire task freenas is doing is basically going through those emulated device drivers: read disk -> send to network and read network -> send to disk. Apparently the overhead is just too high for good performance.

So, it works well for testing. But if you care about performance, not so good. And if you didn't care about performance, you wouldn't be intersted in network attached storage. So for most people I can't recommend it.