2016-02-12, dgastler
It turns out that cms1 is not very fast and is the bottleneck for these speed tests.
Dominique instructed me on how to get the FEROL to drop the data when it lands and now the CPU use has dropped dramatically and the throughput went way up.
Here are the commands to throw away the data:
>lspci -v
02:03.0 Memory controller: CMS DAQ group Device fea0 (rev 01)
Flags: 66MHz, slow devsel
Memory at dd210000 (32-bit, non-prefetchable) [size=64K]
Memory at dd200000 (32-bit, non-prefetchable) [size=64K]
>sudo busybox devmem 0xdd210000 32 0
With 1952 Byte events, I was able to max out the AMC13's random event rate at 130kHz.
To try to max out the link, I increased the event size until I got backpressure. This happened at FAKE_DATA_SIZE of 0x40 (6560bytes) with a max rate of about 75kHz.
This gives a throughput of 500MBps, which is what we would expect. |