[ntp:questions] How bad is USB?

Uwe Klein uwe_klein_habertwedt at t-online.de
Fri May 1 20:07:16 UTC 2009


David J Taylor wrote:
> Garrett Wollman wrote:
> 
>> In article <K9WdnS4y5pIuymfUnZ2dnUVZ_t2dnZ2d at giganews.com>,
>> Richard B. Gilbert <rgilbert88 at comcast.net> wrote:
>>
>>> USB is nearly useless for NTP!  USB has latencies sufficiently large
>>> and variable to render it unsuitable for use with NTP.
> 
> []
> 
> Garrett, Richard,
> 
> You've both commented that USB has drawbacks, but in reality what 
> performance might be obtained?  Not everyone needs microsecond 
> precision, and USB might allow millisecond precision - i.e. possibly 
> better than what might be obtained by Internet access alone, or by using 
> a radio source.
> 
> Has anyone made any actual tests or measurements?  Are there any results 
> available from an actual installation of USB on either Windows or UNIX?

USB is a polled system. i.e. every action is initiated by the
host controller and peripheral "interrupt packets" are just
proper answers to a host enquiry.

The basic rythm of USB is 1ms.

HighSpeed introduces 8 subframes of 125us in this 1ms frame.
But bus signaling is at FullSpeed .
So peripheral responses are rasterized into this 1ms rythm.

Depending on OS you get a plethora of "funnies" added
into latency behaviour. The potentially best behaviour
can be expected from isochronous transfers.

uwe




More information about the questions mailing list