Hi List,
I want to share memory between several instances of PD running on separate reserved cores. I've tried the share-mem lib, but it has a very high overhead.
Since I am writing custom externals for the puredata patches, I have started to include writing the data to be shared into files on a ramdisk. It does work quite nicely, but I am wondering whether some more elegant solution could be done using the Memory-Mapped Files, or are there any other simple and fast solutions around?
Has anyone on this list experience with this? I don't actually need to share large data sets, mostly it's just bytes that indicate the state of the patch. Network based solutions have proven to be too slow.
These patches will run on raspberry pi.
Thanks! S
hello, can you elaborate about share-mem lib overhead? I'm not aware of such problem. cheers Cyrille
Le 18/09/2021 à 18:43, Sebastian Lexer a écrit :
Hi List,
I want to share memory between several instances of PD running on separate reserved cores. I've tried the share-mem lib, but it has a very high overhead.
Since I am writing custom externals for the puredata patches, I have started to include writing the data to be shared into files on a ramdisk. It does work quite nicely, but I am wondering whether some more elegant solution could be done using the Memory-Mapped Files, or are there any other simple and fast solutions around?
Has anyone on this list experience with this? I don't actually need to share large data sets, mostly it's just bytes that indicate the state of the patch. Network based solutions have proven to be too slow.
These patches will run on raspberry pi.
Thanks! S
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list .
indeed I can't!
It has been a quite a while that I had attempted an implementation, and on revisiting it now, I cannot detect any overhead.
Apologies for this wrong accusation and it looks that I can continue with designing the patches using shmem!
Best, Sebastian
cyrille henry wrote on 19/09/2021 10:22:
hello, can you elaborate about share-mem lib overhead? I'm not aware of such problem. cheers Cyrille
Le 18/09/2021 à 18:43, Sebastian Lexer a écrit :
Hi List,
I want to share memory between several instances of PD running on separate reserved cores. I've tried the share-mem lib, but it has a very high overhead.
Since I am writing custom externals for the puredata patches, I have started to include writing the data to be shared into files on a ramdisk. It does work quite nicely, but I am wondering whether some more elegant solution could be done using the Memory-Mapped Files, or are there any other simple and fast solutions around?
Has anyone on this list experience with this? I don't actually need to share large data sets, mostly it's just bytes that indicate the state of the patch. Network based solutions have proven to be too slow.
These patches will run on raspberry pi.
Thanks! S
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list .
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> https://lists.puredata.info/listinfo/pd-list
a couple of loose thoughts, I don't know anything about pi:
- you mean that using [netsend] / [netreceive] is too slow compared to
writing/reading to ramdisk?
- would it be better to centralize the patch+instances using pd~? (I
imagine [clone] uses the same thread)
- if these are single values, send them through virtual midi or audio
channels?
Hi List,
I want to share memory between several instances of PD running on separate reserved cores. I've tried the share-mem lib, but it has a very high overhead.
Since I am writing custom externals for the puredata patches, I have started to include writing the data to be shared into files on a ramdisk. It does work quite nicely, but I am wondering whether some more elegant solution could be done using the Memory-Mapped Files, or are there any other simple and fast solutions around?
Has anyone on this list experience with this? I don't actually need to share large data sets, mostly it's just bytes that indicate the state of the patch. Network based solutions have proven to be too slow.
These patches will run on raspberry pi.
Thanks! S
-
cyrille henry -
João Pais -
Sebastian Lexer