Chris Wright wrote:
>
> What types of embedded platforms are you thinking of, are these platforms
> going to have any way to install this new application? For example,
> something like Network Attached Storage doing NFS...sure they support RPC as
> part of NFS, but there is no way to add another RPC server to the NAS device
> itself.
This is where I'm hoping that the openness of the specification will work--
the NAS vendors will put the RPC or whatever service in place, because they
want to sell cluster storage. It works for everybody if we can agree on the
messages and the protocol. Working some things out in this linux-centric list is
just the first step, if it is going to work out.
>
> Somehow, it seems to me that the _actual_ messages being passed via the
> GRITS verbs are more important than the transport protocol.
Yes, but getting some sense of the supportable protocols will influence
our thinking, so getting short-list seems like a good idea to me.
> All I can say
> is... do we really want to do our own data marshalling?
Amen! That's why I'm down on raw tcp, despite my colleagues, who I will
have to keep talking to. System guys by and large still don't trust this
new-fangled RPC business, and like to do things on their own.
-dB (one-time CORBA ORB implementor)
>
> What types of embedded platforms are you thinking of, are these platforms
> going to have any way to install this new application? For example,
> something like Network Attached Storage doing NFS...sure they support RPC as
> part of NFS, but there is no way to add another RPC server to the NAS device
> itself.
This is where I'm hoping that the openness of the specification will work--
the NAS vendors will put the RPC or whatever service in place, because they
want to sell cluster storage. It works for everybody if we can agree on the
messages and the protocol. Working some things out in this linux-centric list is
just the first step, if it is going to work out.
>
> Somehow, it seems to me that the _actual_ messages being passed via the
> GRITS verbs are more important than the transport protocol.
Yes, but getting some sense of the supportable protocols will influence
our thinking, so getting short-list seems like a good idea to me.
> All I can say
> is... do we really want to do our own data marshalling?
Amen! That's why I'm down on raw tcp, despite my colleagues, who I will
have to keep talking to. System guys by and large still don't trust this
new-fangled RPC business, and like to do things on their own.
-dB (one-time CORBA ORB implementor)