Looking through the source code of udpsend~ I don't think broadcast sends a different packet, I'll keep looking.
Also note that multicast is supported. Have you tried that? It should be more efficient than broadcast.
(udpsend-help has the messages used by [udpsend~] for multicast, udpsend~-help needs updating).

Martin


On 2014-07-24 14:46, Bektur wrote:

Thanks for the advice!

I’m sorry for the late reply, but I finally got my hands on a windows machine, and here’s what I got:

- The issue with bad header tag in case of a broadcast is still there, so at least it’s not related to platform-specific implementation
- I managed to make it work by creating a set of separate [udpreceive~] objects for each client and they work almost without any noticeable problems with sound

I wish I could use wired connection, but since my clients are mobile devices, so I have to go with Wi-Fi.

Just in case if someone else would be looking for this problem, I think jack.udp is another possible option (since IIRC netjack and jackport don’t have mobile implementations). However, for now libpd + [udpreceive~]/[udpsend~] work well in local wireless networks.

Best regards,
Bektur

On Jul 21, 2014, at 11:06 PM, Martin Peach <martin.peach@sympatico.ca> wrote:

On 2014-07-21 12:50, Bektur wrote:

I’m running a simple iOS application with libpd and udpreceive~ objects in the patch. The audio is sent from a computer with udpsend~ object, with only one channel currently being used. Whenever I stream to a single IP address everything works fine with almost unnoticeable latency. However, when I try to broadcast to multiple addresses in the network (as shown in udpsend~ help patch), I get two kinds of errors:

– On OS X, block size >= 512 triggers “Message too long (40)” error. It happens only on OS X (tried the same patch on ubuntu, and this error didn’t reappear), one of the possible solutions I found was to use smaller block size, but in such case it would cause the second kind of error

– On a client I get "udpreceive~: bad header tag (1)” error with incrementing numbers in the error. At first, I thought the libpd implementation was to blame, but it also happened when I run the same send and receive patches on two different computers running Pd-extended and connected over a local wireless network.

I haven’t found any workaround for this yet, but I’m planning to try reproducing these issues on a windows machine with ASIO drivers.

I suppose if you do it over a wired connection it will work better. Probably the WiFi is dropping packets and retransmitting, so things get out of sync. I don't know how broadcast UDP is implemented in WiFi, but I imagine it involves sending copies of the same packet to each receiver rather than a single packet addressed to multiple receivers.

Martin