[nanomsg] Re: weird extra byte in header for IPC
- From: Martin Sustrik <sustrik@xxxxxxxxxx>
- To: nanomsg@xxxxxxxxxxxxx
- Date: Mon, 21 Apr 2014 10:01:38 +0200
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On 20/04/14 19:49, Garrett D'Amore wrote:
> So I understand the idea. But I think this feels like a different
> protocol than the IPC we have today. I'd argue this is like some
> new shmem protocol. It might even be possible to use some form of
> rdma for such in the future to allow it to operate outside of a
> singe host. For now I would have preferred to keep that extra
> byte out and invent the new transport later. With a different URL
> scheme as well.
This is an interesting topic, in fact.
The underlying principle is that we want to expose existing
functionality of the transport protocol, rather than trying to augment
or normalise it. (I think this idea was already touched upon in the
UDP transport discussion earlier on.)
In practical terms, this means:
1. TCP has no support for true out-of-band transport like RDMA (TCP's
OOB feature is not really a transport mechanism).
2. Possible IB (or IB verbs) transport would have support for RDMA,
optimisation for transporting small message in-band etc.
3. IPC transport has support for classic in-band transfer as well as
passing the shmem descriptors.
It's not very elegant or very consistent but it's something we inherit
from L4 and is not our job to fix IMO.
Martin
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/
iQEcBAEBAgAGBQJTVNBiAAoJENTpVjxCNN9YNhoH/1+8hHKXYDmqEDvRWe+lGA9v
tx29ZOxQJMDOaCoUA9sjxjWwcOZhZdnWU8T6YjHm93EApkIz4tguWfPOIIBeHemB
yuIHwaiAYrdMEI4FF/QoOjGe5Zegn2UyEADyjKNbJ69Kj5VdSTeEpipKNLtDF6ld
+QX+imiiswczJ1xTzVLo7C9aO2UChXQtLp6LH44ljIGoZkwh+ViqqxTq39LFF2du
a/ddENSaDCE7zWWyxTSVVOwVyfhL5YIVk+GxK4xagPu+ekL4uQt8TQJ84Qaa2nJx
QAk1kGqjqlbUO39Qzg3NnJwUViEztOAA8bNZuZ4WP0u2bvFC5lJHJziD6bc5WUY=
=zX22
-----END PGP SIGNATURE-----
Other related posts:
