greetings.

I am investigating using netsend to send patch data from one pd instance to another, to make use of both cores of my cpu. I have replaced most of my internal patch wiring with [send] [receive] objects. is it possible to [send] to a [netsend] and [recieve] from a [netrecieve] or am i fundamentally mis understanding what these objects do?

I am trying to shuttle about 20 streams of data to another instance. how should i go about formatting these individual streams (which are currently [send] objects) so that they reach their intended destinations.

thank you for any insight

Onyx
www.indiegogo.com/beatjazsystem

or...should i possibly create a new netsend for each existing send object?

You probably don't want to be sending 20 streams at the same time, unless they are updating rarely. I would go the route of serializing, or creating lists of parameters and then sending that list on a "heartbeat" signal.

[netsend] and [netreceive] are quite different to [send] and [receive], [netsend] sends messages across a network. In your case it sounds like you'll be sending to your computer's own IP or 'localhost'. Then use a port number that isn't currently in use.

You could think of [send] and [receive] as local to a single Pd instance and [netsend] and [netreceive] as a way of communicating between multiple instances.

thank you for the reply.

how would i go about sending 20+ streams of data simultaneously over a network connection? most of the streams will be updating alot and are very time sensitive.

thank you

onyx

Try it. Use UDP, though. TCP/IP validation will surely slow you down. I would oopen one netsend connection, and attach it to a global receive, then send to that receive whatever you need send out over the socket. Use messages as in the example: [send FOO mymessage[ Then, use a netreceive on the other end, piped into a route to get the messages. If you need help, I can whip up a patch for you.

This is what I was thinking. My netsend isn't working properly, though. Proceed at your own risk.

http://www.pdpatchrepo.info/hurleur/netsend_rec.zip

"you are a god amongst insects..."-Magneto

thank you! that was it. specifically, your patch illustrated that the data should goto a [1 $1< message, then to a send object. then the receive object feeds into a 8send $1< message. then a [route] object on the netreceive side. i was attempting to use [pack][unpack] on both ends, unsuccessfully. your method works perfectly! thank you for your insight!!

onyx

ooops...wait a sec. my switch triggers are working properly but the accelerometer signals are doing anything. no variable streams are working. what am i missing?

thanks in advance

onyx

ok...a little research yielded using

into in place of the [send $1< message. everything works and my cpu dropped from 85% to like 15%!! found that putting the gemwin on the second monitor, upped the cpu usage by like 30%. fyi. thanks again.

Awesome. Wanna post your patch?
Are you using linux? How are you launching the two Pd instances?

here it is. it works but i havent had to use it as i decided to get my patches working in linux and man! what a difference!! everything runs perfectly without all the work arounds and hiccups! i just open multiple instances and they just work. I'm sure they are still drawing off the same core but they work so much better that its not a problem. I am satisfied. i will use the netsend patch to link with the synth abstractions i start on later in the week. thanks again for the help!!

Onyx

http://www.pdpatchrepo.info/hurleur/Beatjazznetsend.pd

Hi guys,
just a question: when you use [netsend]+[netreceive] (or [pd~]) on the same computer - but in two separate instances of PD - does automatically PD optimize the dual-instance for a dual-core processor, or I have to set something?

I believe one of the original purposes of [pd~] was to make sure you could run another instance on a separate core with a single patch.

@adrjork said:

Hi guys,
just a question: when you use [netsend]+[netreceive] (or [pd~]) on the same computer - but in two separate instances of PD - does automatically PD optimize the dual-instance for a dual-core processor, or I have to set something?

Your operating system will manage determining which core a given thread runs on for you. It may even decide to run them both on the same core, if the other cores are busy with other things. Sometimes it may even decide to switch threads from one core to another while the thread is running.

E.g. assuming Maelstorm is correct about pd~ (I'm not familiar with it), it actually runs on a separate thread, not a separate core. It may or may not run on a separate core if that's what the system decides makes the most sense.

Also keep in mind that PD in GUI mode has other things going on, and is already running multiple threads. When multiple instances of PD are running your OS will distribute PD's various calculation and GUI threads, as well as the threads of other processes running on your system, among cores as it sees fit. Your level of explicit control over this decision is very small.

You can set affinity masks for entire processes through the task manager in Windows, not sure if/how you can do it in OSX/Linux, but you are unlikely to make a better decision than the OS here.

You are better to not ask questions like that unless you observe that PD is running too slowly for your requirements and have determined already that spreading processing out on different cores will solve your bottleneck.