Diretta Audio Transport Protocol

[Qhwoeprktiyns]

Just read about this new protocol being developed in Japan: https://www.diretta.link/top.html

 

Has anyone heard of this and understands how it works ?

[mansr]

[Qhwoeprktiyns]

Thanks for your contribution. 

[YU]

Flattening target process  by small send data size and short between time.
 

 

[Qhwoeprktiyns]

Thanks. So what is the theoretical benefit? Less processing on the client side? 

[YU]

Target processing is small.
Load fluctuation to target power supply is small

[Miska]

1 hour ago, YU said:

Flattening target process  by small send data size and short between time.

 

Smaller the data block, higher the protocol overhead and CPU load...

 

Quote

Thanks. So what is the theoretical benefit? Less processing on the client side?

 

Smaller the packet, more processing time it takes on the client side because of involved overheads.

 

Quote

Target processing is small.
Load fluctuation to target power supply is small

 

And in worst case you send packets so frequently that the packet rate falls into audio band. Like UAC's 8 kHz... Then it shows up easily in the measurements...

 

 

P.S. Is this yet another protocol that doesn't work over WiFi? (like AES67/Ravenna/Dante/AVB)

[YU]

That's right.
If the packet size under the MTU size, processing high.
multiple packets can not be processed at once,then  tasklet occur frequently.
but processing per one transaction is short process time.

Diretta is  optimize best MTU size.  
Target do not manage buffer control.
designed to reduce processing, use UDP etc.

case
22Mhz DSD      Send Sycle = 500usec  size = 1494 byte
44.1Khz PCM   Send Sycle = 2000usec(limit) size = 767 byte

[Miska]

6 minutes ago, YU said:

That's right.
If the packet size under the MTU size, processing high.
multiple packets can not be processed at once,then  tasklet occur frequently.

Diretta is  optimize best MTU size.  
Target do not manage buffer control.
designed to reduce processing, use UDP etc.

 

There are good zero-copy TCP offload engines that handle large DMA transmissions without CPU involvement...

 

Just do full speed transmission of a file over 1 Gbps or 10 Gbps ethernet and observe which method has lowest CPU load...

 

This is quite well optimized, since practically entire Internet relies on fast transmission of huge amounts of data over TCP.

 

[YU]

It has a different purpose.

TCP need send ACK.....
designed to suppress the effect on the target.
Only use of the local network.
It is not aimed at increasing the throughput.

We are planning to implement FPGA and Hardware.
ZeroCopy is a future task

[Qhwoeprktiyns]

I assume you are from Megatech, in Tokyo? If so, can you tell us more about the release dates? 

I understand there are two components: sink and synch. Are both available under Linux? 

[YU]

Yes we had linux liblary.(company license)
we will release  binary on ALSA Sink and Sync after one month

[Qhwoeprktiyns]

Thanks. Perhaps you could let us know in this thread? I am sure many will be interested in testing this. 

[Miska]

1 hour ago, YU said:

case
22Mhz DSD      Send Sycle = 500usec  size = 1494 byte
44.1Khz PCM   Send Sycle = 2000usec(limit) size = 767 byte

 

Now the challenge with this is similar to USB Audio Class...

 

500 µs = 2 kHz

2000 µs = 500 Hz

 

While USB Audio Class uses 125 µs = 8 kHz

 

These fall into audio band and can be frequently detected from the DAC's analog output.