jueves, agosto 04, 2011

Building my own NAS for DLNA (III): performance

After introducing the test in part I, and comparing the specs of the two contenders in part II, let's do some benchmarks.

Test files

When I was setting up this test, I needed to use some HD big files, like on my very own usage. Some googling, and I found a 20 Mbps, MKV contained HD-DVD file (278.3 MB) and a fantastic collection of HD trailers from where I chose Sony's "Paint" (72.9 MB), one of the first I saw when this LCD era started. These two files add up to 352.64 MB.

I fancy watching Top Gear. In fact, it's the only TV show I wait for all week, and when a season ends... well, I can't wait for it to resume!!. So I have chosen a couple of episodes: Season 16, Episode 05 and Season 17, Episode 06, both 720p and about 1.45 GB each. I recommend you to use some kind of video you know in detail when testing, to see any difference. If you can almost say out loud the script of Matrix, that's the way to go! :-)

UPDATE: for those trying to absolutely squeeze your AV equipment, I recommend Samsung's "Oceanic Life": 40 Mbps of bitrate.

Network performance

I'm using my main PC and a Comtrend wifi router, both of them having gigabit ports. The problem is that the candidates have only Fast Ethernet ports, so we shouldn't expect lighting fast network speeds. In fact, the first thing to find out is raw network speed, without testing the hard drives. Jperf is a frontend for the Java based Iperf network speed app. It only needs to have Java JRE 1.5+ installed, and runs in server-client mode (you choose one of them) to max out the available bandwidth between them.

mmm, not very quick...

In fact, the differences are very small. The slowest one is Windows XP running on the Travelmate achieving 9.4 MB/s; switching to Windows Server 2003 makes it 8% faster, while the newer Aspire One is 12% faster. I told you, having 100Mbps interfaces won't make them very quick if you plan to copy very very big files.

I then time how long does it take to copy FROM (read test) and TO (write test) the "NAS". DiskBench is a free program that comes in handy. I copy twice one of Top Gear's episode, about 1.45 GB:

DiskBench is not a tool designed for everyday copying.

Mmm, it seems to me like the network limit is very close the machine's disks bottleneck. Travelmate wins when running Win 2K3 when reading, but it's always the loser when running Win XP.

I decided to do some more real life tests, as I'm not usually using DiskBenck for file copying. It's been years since I discovered CopyHandler, a replacement for the stock copying capabilities of Windows (cancel, resume, queueing and many more). Perhaps not the best of its class, but still very useful. It lets you even change the size of the buffer used when copying, so you can play a bit to adjust this parameter to fit your scenario. I copied over the 2 HD small videos (352.64 MB) three times, the average results are in MB/minute:

CopyHandler is a nice tool to use.

Increasing the buffer size from 128 kb to 2 Mb seems to work very well in all cases: when reading TravelMate under 2K3 is 10% faster and Aspire One 16%; when writing, TravelMate XP is about 5% and Aspire One about 12%.

Let's say we can achieve around a 500 MB/min performance with the TravelMate and about 450 MB/min with the Aspire One.

Streaming content

Then I started streaming content to see how much bandwidth is needed. It doesn't make any difference, as the results are almost the same while viewing one of the Top Gear's episode. You can monitor the network usage with the Performance Monitor in Windows, but to record average and max values, I've chosen another free little app, SoftPerfect NetWorks. You can even install the taskbar addon to monitor live bandwidth or start its speed meter to view max and average values:

  • streaming (average): 600KB/s
  • streaming (max): 1,55 MB/s
  • streaming (initial maximun): 6,15 MB/s
  • FF/RW skipping: 10 MB/s

The network usage is maximun when skipping the video (10 MB/s), up to the network's limit in my case as shown by JPerf. In fact, I noticed that skipping is faster when viewing the files using a directly attached USB stick. The second most demanding event is the start of the streaming (6,15 MB/s) perhaps while the player is buffering some seconds. After that, the average network usage is about 600 KB/s with some peaks at some scenes.

I think these results show that these machines are good enough to keep HD content streaming, as most of the hard work by Serviio is done when scanning the files, not while streaming them. We will see it in the next chapter (power consumption and CPU usage).

Building my own NAS for DLNA (I)
Building my own NAS for DLNA (II): specs
Building my own NAS for DLNA (III): performance
Building my own NAS for DLNA (IV): CPU usage and power consumption
Building my own NAS for DLNA (V): Conclusions

Etiquetas: , , , ,

Si te hemos sido útiles, ¿por qué no demostrar tu agradecimiento? Haz clic en alguno de nuestros anuncios, nos ayudarás a seguir con artículos como este.

También te recomendamos que visites nuestra página "hermana". La temática es un poco diferente pero seguro que también a tí o a tus conocidos les resulta de interés:

Related Posts Plugin for WordPress, Blogger...