[nanomsg] Re: weird extra byte in header for IPC
- From: Martin Sustrik <sustrik@xxxxxxxxxx>
- To: nanomsg@xxxxxxxxxxxxx
- Date: Sun, 20 Apr 2014 13:33:41 +0200
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Hi Garrett,
> You can't make that work generally I think.
>
> You'd have to have another region of shared memory established.
> That feels like it should be a totally different transport.
> Indeed with shared memory you would really like to establish a
> protocol for placing data in the shared memory regions.
I was following the system often used in high-perf networking: The
actual memory managements mechanism is determined based on the message
size. Typically, you get three different levels here:
1. Very small messages - don't even attempt to do zero-copy.
2. Middle-sized messages - allocate the message within a pre-allocated
dedicated pinned-down memory chunk.
3. Large messages - allocte a new pinned-down memory chunk per message.
Anyway, the point is that all 3 compose a single transport mechanism,
and the actual thresholds to switch between the different allocation
options is just a policy to apply to the transport.
> Anyway nanomsg doesn't have anything like this at present.
True, but the protocol, hopefully, should out-live the current
implementation.
Martin
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/
iQEcBAEBAgAGBQJTU7CVAAoJENTpVjxCNN9YodIH/2904BrBw5s+v3IM2zGnfZVf
IaGQOhxiaBlqq/9Cjn5zRu9tf47ydEoKEYIWjvl2iqsoQS4ZIgwhcQe9zypX4rlc
aHcs8egJcYxMIkiIPFIREb2AlsrxSXQdkLzVWu3FxWNVGqwroHT3JEyfJEooCmV/
DjH85nYffi2ygStT3BjT1claLwYS/P8ReJz7HDLPl+wo0oQYP481VpYmbihlM82q
1niIOGRCehtJ3/D3035tns/hxbU7HqmhkYTRFwc4DER3ALOXyh8dATtdTCAsTeHz
pXeub43j1uLU9bzSydjzoWC8cLxsU8l9KbpdNFSd7UnnkC0USRghl2E2Vl/qPVo=
=61JI
-----END PGP SIGNATURE-----
Other related posts:
