usb DAC bandwidth utilization during playback

[ricko01]

I have a client/server topology using LMS (server)/squeezelite (client) for music playback and nfs (server)/kodi(client) for music video playback.

 

Because everything matters, I have looked at ensuring that the USB bus used for the USB connection into my DAC from the client PC is reserved only for audio traffic.

 

As part of this, I drilled down into how much bandwidth is actually used... so thought I might share this.

 

Ignoring the baseline hub devices, I have the four active USB devices:

 

- the USB stick I boot ramroot off

- the port that goes to the DAC

- a USB wifi dongle (used to allow me to 100% control network bus utilization)

- a USB dongle for mouse/keyboard

 

The bus allocation is as below:

 

---->Bus 001 Device 003: ID 0951:1666 Kingston Technology DataTraveler G4         [BOOT USB]
---->Bus 001 Device 004: ID 20b1:2017 XMOS Ltd                                    [DAC USB]

 

---->Bus 002 Device 003: ID 0cf3:9271 Atheros Communications, Inc. AR9271 802.11n [USB WIFI]
---->Bus 002 Device 004: ID 046d:c52b Logitech, Inc. Unifying Receiver            [KEYBOAD/MOUSE]

 

With the system "idle", I saw constant loads. Interestingly the keyboard/mouse dongle has no traffic when not being used:

 

- BOOT USB      does 0 KiB per second
- KEYBOAD/MOUSE does 0 KiB per second

- DAC USB       does 4 KiB per second for both send and receive
- USB WIFI      does 20 KiB per second for both send and receive

 

When moving the mouse, total bandwidth used (send+receive) was 10 - 20 KiB/s

 

To test my worst case utilization I streamed a music DVD with a 16bit/48k sound track (worst case cause the network traffic includes both video and audio)

 

Graphing the network load over a 15 min period we see cyclic transfers peaking at ~15MB/s (testing squeezelite shows similar cyclic transfers but peaks were ~11MB/s for ~4 seconds at the start of each track)

 

Looking at the USB DAC traffic for the 16bit/48k sound track we see basically send and receive in a constant narrow band of utilization (to the DAC a constant 650 KiB/s and from the DAC a constant 300 KiB/s)

 

For my purposes, this analysis showed the placing of the network dongle and USB DAC cable on separate ports did in fact provide separation and also confirmed that during playback the other devices that co-exist on the same buses as the network dongle and USB DAC use zero bandwidth.

 

It was interesting to get metrics on actual bandwidth utilization as part of the analysis.

 

Peter

 

 

 

 

[cjf]

FWIW, USB 2.0 is said to max out around 60 (MBps) and USB 3.0 is said to max out around 640 (MBps) both of which are waaay beyond what any audio system is likely to ever ask for under any condition.

 

I believe there is somewhat of a standard in place in packet size for USB when streaming audio that is around 1MB chunks but depending on the software used you may be able to tweak that number. As an example, if memory serves me correctly, MPD Voyage Linux allows you to alter these values but it may have no useful outcome since the USB device itself usually tells the OS what it is capable of receiving in terms of data packets via the driver.

 

I did a similar series of tests back in the day for Ethernet streaming via hard wire/wireless to determine what connection speed was required based on various Bit/Sample Rate files being streamed. This was mostly done out of curiosity for wireless bandwidth requirements.

 

Here is a link to that Blog post if your interested in seeing the info. Unfortunately the pics/graphics have been stripped out over the years but most of the content is still intact. It appears I can't edit that Blog anymore otherwise I would add the graphics back.