Thecus N4800 NAS Reviewed

[holdem]

Hello!

Having just bought an N4800 I was very interested to see the performance graphs in your review. I bought the 4800 to replace an old, slow NAS and I wanted to store a lot of video and lossless audio so the files are large. I'd already bought and populated it before I found your review but you give it a good review so that's good, heh.

My 4800 now has 4 x 3TB Seagate ST3000DM001-9YN1 in unencrypted raid 5. They're on the compatibility list (minus the -9YN1 part) and are all listed as healthy. They all pass the short test and they list as SATA 3Gb/s even though I think they're actually 6 but I assume that's because the 4800 is only 3. On the whole I like the 4800 so far. Good interface, easy RAID/network setup, nice neat little quiet box even with 4 drives twirling merrily. Pretty much it all just worked straight away.

My post stems from the fact that you seem to get a much better transfer to and from the 4800 than I can get. I'm not a network storage whizz by any stretch so I'm not entirely sure what constitutes "good" performance. It's also equally possible I'm just misreading your charts!

I mentioned the unencrypted bit above as at first I thought that was my problem. When I initially created the array I stuck a memory stick in the front and encrypted the array, mostly because I could to be honest. I figured there'd be a performance hit but was curious how much it was. I performed a hugely unscientific test of copying a single large (19.1 GB) file onto it and it capped out about 25 MB/s. This is better than my old NAS by far (don't ask) so I wasn't displeased. I then deleted the array and recreated it as unencrypted for comparison. That capped about 35 MB/s. This is single file write from an Ubuntu 10.10 desktop over SMB, reading from flash. Copying multiple (semi-large 25-30 MB) files from flash onto it give the same figure. Reading back the same file from the NAS caps about 41 MB/s, writing to magnetic (my flash drive is full). Both (desktop and NAS) network adapters are listed as 1000 Mb/s.

I stuck a crossover cable in LAN2 direct to the desktop and got the same figures so it's probably not my cheap switch. I'd already had cable problems using a 5e that dropped the network to 100 so all the cables are cat 6.

As I said, I'm not displeased with the 4800 at all - it beats my old NAS into a cocked hat so it's all good. I'm just curious though why my figures are seemingly so far from your graphs. You have figures of 90 MB/s upwards yet I can't break 45!

Any fault finding or tuning tips gratefully accepted. I'm not a network engineer so I'm open to the fact that my methodology is rubbish and I've misread your graphs!

edit: Sorry, filesystem is EXT4.

 

[thiggins]

I suspect your Flash drive is limiting performance. We run filecopy tests from two Samsung Spinpoint ES drives configured in RAID 0.

 

[holdem]

Sorry, when I say flash I was meaning SSD. I'm not talking about a stick, that was only when I was encrypting the array to store the key. Apologies if that wasn't clear.

If I run the disk utility and run the read-only benchmark on that SSD, it gives a minimum read rate of 283.4 MB/s. My other SSD shows a minimum read rate of 251.9 MB/s and the 500 gig magnetic caps at 60 MB/s read which I hit before in previous in-machine copies between those 2 drives. I've also managed ~64 MB/s across the same switch to another linux box so I'm reasonably comfortable the switch is OK.

I have another machine with an on-board nvidia LAN using a forcedeth driver which manages 45 MB/s read, 43 MB/s write from/to the 4800. I also have a 3rd machine with a marvell/sky2 LAN which manages 49 MB/s read, 45 MB/s write from/to it. The machine I spoke about originally has an on-board Realtek 8169. Given that this is the machine that managed 60 MB/s across the LAN to another linux box before, I'm curious where the fault lies. None of these machines can get anywhere near your benchmark figures, even for read.

I'm semi at a loss where to start digging really. It's a shame I can't get an SSD into my "good" machine, the one that reads at 49 MB/s. Even at 49, it's still only managing half your read figures. I suspect that machine is write capped by it's magnetic, it would be interesting to see if it went up with an SSD.

I'll clear some space on an SSD and see what I get copying back from the 4800.

 

[holdem]

OK.

Copying the same 19.1 GB file back to SSD gives 45.6 MB/s read. I'm inclined to point my suspicious finger at the on-board Realtek but don't have a "real" adapter to replace it with. That might be a good purchase. I can probably believe the other 2 machines are write capped by magnetics - neither of them are blisteringly fast machines. Not sure I've tried copying back onto the 2TB magnetics in one of them; I might try that next.

I'm assuming here that the NAS is doing what it's supposed to do and I don't think I have a way to prove it's not with the kit I have. I can see no obvious settings I've missed but as I said, I'm not really a network whizz.

Interestingly, copying the 19.1 GB file back to SSD from an encrypted 4 disk raid 5 array yields 35 MB/s read.

All the CPU graphs on the 4800 seem to say 0 (zero) all the time. The network ones go up and down as does memory usage, but CPU always shows nada.

 

[holdem]

Intel x1 PCI adapter gives approx 25% increase over onboard 8169.

107.5 MB/s FTP to unencrypted 4 disk raid 5 array.

Samba - Bah.

Also, the scheduled power off/on facility forgets the key for encrypted partitions so you have to re-insert after power on. If you boot without the key in, there also doesn't seem to be a way to get it to recognise the array without inserting the key *and* rebooting. Hmm.