Raspberry Pi Stack Exchange is a question and answer site for users and developers of hardware and software for Raspberry Pi. It only takes a minute to sign up.
Sign up to join this community
I'm currently using the Pi overclocked to 1GHz on Archlinux running deluge and samba which access a 3TB external USB drive.
I plan on buying another identical drive and set it up as a raid 0 configuration (no, I don't care about the data being lost).
Would this work OK on the Pi? And if yes, is there likely to be an influence on speed of say, transfer rates on samba if I do this, or does it not matter because the bottleneck is the CPU anyway?
I love the Pi, but I have to tell you it doesn't hold the fastest IO in the market.
According to this article on the USB performance there is a empiric limit of 30MB/s transfer rate on the USB port. And all the USB ports use the same IO interface. As that, the limit of all combined USB should be capped on that 30MB/s limit.
If you take a look at some HD throughput charts you will see that the average HDD can pull much more than those 30MB/s, so, seems obvious that the limit on the port will keep transfer speeds much below the empiric maximum of most HD Drives.
Given that the limitation is the port itself, connecting 2 HD will only make them share the available throughput, and not add to any speed.
Overclocking is not great and causes allot of kernel panics under high load. Also the biggest bottleneck will be the USB controller hammering drives and serving data over LAN.
You could offload the Software raid onto a nice USD Raid controller. I am not sure exact models of your external drives but if they have eSATA you can use a cheap converter cable to connect them to this. This will give you maximum Hardware RAID performance, offload the CPU and only USB bandwidth for data and LAN, maximising read speed.
One problem with RAID0 is that it doesn't really double speeds on soft or cheap hard raids, its actually pretty rubbish. This is why ZFS was made and its supported on Linux ARM. The only problem is RAM and it may run out kernel address space if hammered, but with tweaking you will get MUCH better read and write perfomance on USB attached drives. This is because ZFS can read and write to each drive seperatly using RAM cache (this is why RAM is essential) so parts of files stripped across drives are read at full speed from each drive, assembled and served at a low level. So if you really want performance you should use a micro server/ PC with 4GB (3TBX3TB) but more is better. The results for software spanned drive pooling will blast you away. I use 4X1TB @ 4GB with ZFS on my HP Microserver and its mint! Copying from the server saturates my 100mb LAN and these are cheap, mixed set drives
The PcDuino3 A20 is a great alternative to the Pi.
Its got 1GB RAM, Gifagit LAN, Wireless and a SATA Port, where you can attach a port multiplier to connect a whole bunch of Hard drives in RAID. Its great because you can install the full Ubuntu onto it or Android 4.2 possibly even something like RouterOSEverything but ARM? or pfSenseDoes not support ARM :( but FreeBSD is kind of supported which can do all that.
Hmm, I got much faster reads from 4x USB drives as a raid than a single.
I currently have a raid 0 of old HDD's running MySQL server at the moment, and it moves through data pretty quickly. For some reason writes seem a bit slower than I expected. But reads are PDQ.
I haven't done any side by side tests though.
