/dev/random throughput testing requested

Development
1

I want to get a an idea of the rng throughput on machines different than mine so I can have an idea on a seinsible default to rate limit virtio-rng’s access to a guest will be.

The current limit of 1 kb/s is extremely low bordering on useless.

Note that you don’t have to be a Linux baremetal user to try this out.

Test:

  • Run: dd if=/dev/random bs=1k count=1 for 5 times then average the results

  • Please do not run any crypto intensive operations on your machine during this test as it will skew results because some of the rng input will be used up.

2

On baremetal Debian-8,

506 kB/s
69.0 kB/s
0.0 kB/s
0.0 kB/s
0.0 kB/s

Seemed strange so I looked it up. Something to do with limited “entropy pool”. [ironically, my forum name doesn’t indicate any expertise.]

stackexchange
stackoverflow
superuser

Results from Qubes/Xen VM were also highly volatile but fluctuated in both directions.

Qubes/Xen VM, 1 vcpu, Whonix-Workstation:
141 kB/s
969 kB/s
77 kB/s
89 kB/s
92 kB/s

2nd Qubes/Xen VM, 1 vcpu, Whonix-Workstation:
41.7 kB/s
82.5 kB/s
73 kB/s
97 kB/s
41 kB/s

Also strange, it seems Whonix-Workstation never runs out of entropy.

3

Thanks for testing. The 0 kb/s means the pool is blocking because entropy has been used up and its waiting for new random bytes to generate.

Inside VMs I see something similar here. The VM entropy seems higher and non exhaustible. This is probably the result of running on virtual hardware. Only Dom0 readings would be realistic I think.

4

###Qubes R3.2/Whonix 13

####Whonix-WS with haveged:

250 kB/s
193 kB/s
233 kB/s
229 kB/s
245 kB/s

####Whonix-WS without haveged:

244 kB/s
89.3 kB/s
0.0 kB/s
0.0 kB/s
0.0 kB/s

####Dom0 without haveged:

2.2 MB/s
2.2 MB/s
2.2 MB/s
1.9 MB/s
1.9 MB/s