Re: [Fwd: [Forces-protocol] Presentation of the options forLFB-level multicast]

----- Original Message -----

From: Robert Haas

Wang,Weiming wrote:

In some way, you may call it a thought of relative path. My view is actually
from following thought:
1. A single path is not enough for an operation. In some cases, we may need to
assign several paths for an single operation, e.g.,   we may at the same time
need to set value to 1.2.3 and 1.2.4 where 3 and 4 are field IDs for the same
table. We may also possibly need to set 1.1 in the same operation.

So why don't we support multiple times "path+data" under the same LFBSelect ?

[Weiming]I think we have. Only mutimle 'path, data' in one OpTLV is not sured.

2. It's quite unnecessary we may set different Attributes in one single
operation, therefore the AttrID part is always the same.

I am not sure about this.

[Weiming]Actually my imagine of a PDU format looks like:

OpTLV = AttributeID <DataTLV>

DataTLV = <subpath-value pare>+

Of course, I also don't oppose the format as:

OpTLV = AttributeID <DataTLV> AttributeID <Data TLV>

DataTLV = <subpath-value pare>+

What I really mean is, for a group of subpaths, the AttributeID is the same. Pls see my reply later soon to Joel on this necessity.

Based on above, I think that:
1. It may be very complex to have one specific 'path'  in one 'path' format to
express the actual paths like 1.2.3 and 1.2.4 and 1.1 simulteneously , but it
would be much simpler to have Data field to express it, like the data format as:
    subpath(1.2.3) value, subpath(1.2.4) value, subpath(1.1) value

Instead, take these subpaths out of the data part, and make normal paths out of them. That's doable if we decide to support multiple "path+data" under the same LFBSelect.

[Weiming]I'm afraid you mean LFB Select here OpTLV, is't it?

2. for all subpaths, the head AttrID part is the same, and this is also the only
part that are the same.

Therefore, my thought is the AttrID part is defined in the protocol, leaving the
subpath go along with the data.

I would prefer to have the whole path in the "path", not mixing the subpath and the data. That sounds to me like a cleaner design. Note that this was the consensus in the room after the presentation. But if you have other arguments that would favor the "subpath going into the data", please share them with us.
[Weiming]Obviously to have AttributeID out meets the tree data structure, just the same as we pull out LFB class and LFB instance ID out of the while path. It also makes the protocol very clear in structure, i.e., to set some 'attribute' in some ' LFB insatnce' with some 'LFB class'. While to let a whole path meets the linear data structure, its lower in efficiency and also not very clear in semantics. And also, very offten you will find some part of the path (especially the attribute ID part) is repeating.

Hi Robert, Thank you very much to bring the slides to the meeting. ----- Original Message ----- From: Robert Haas Subject: Re: [Fwd: [Forces-protocol] Presentation of the options for LFB-level multicast] All, I presented Weiming's slides just after Jamal's presentation yesterday. No divergence of views on the principle of how to describe paths was found. Whereas, according to his slides, Weiming considers that the distinction of Attribute, field, and index, must be reflected in the path notation, the consensus in the room was that this is not necessary: a path could be x.y.z, where it is clear that x must be an attribute, and y and z can be field or index. No need to mention it explicitely in the path notation. [Weiming] Actually this is not the key point. While I'm just a little afraid it may lead to ambiguity if , e.g., z can be a field ID or a subscript without tag to indicate it. The path can be constructed with index-search or content-search. The consensus in the room was that the path should include the whole thing, not only the first attribute, as opposed to Weiming's suggestion on the last slide. [Weiming]This is really the key point. We need to verify if it is possible for a single 'path' format to describe all need for path. I just think that, apart from the attribute ID part, others are tightly combined with Data. We may feel difficulty to try to separate path explicitly. Content-search remains to be defined more precisely, as well as block access. So it is too early to disagree ;-) Regards, -Robert Thank you again. Weiming Wang,Weiming wrote: > Jamal, > > ----- Original Message ----- > From: "Jamal Hadi Salim" <hadi at znyx.com> > > > > On Mon, 2004-11-08 at 00:12, Wang,Weiming wrote: > > > > > Jamal, > > > ----- Original Message ----- > > > From: "Jamal Hadi Salim" <hadi at znyx.com> > > > To: "Weiming Wang" <wmwang at mail.hzic.edu.cn> > > > > > > > > > > I still dont see what where we have differences. If Robert

> > > > > individual > > > > presentation :) > > > > > > > The first step is to understand what you are trying to show. > > Look at how many emails it took for you to say "i see the

> it. That's just the 'i see the difference' mean. > > Cheers, > Weiming > > > > So i am not trying to stop your presentation rather trying to

> > see any difference. This includes Robert. > > > > > > > > > -- Robert Haas IBM Zurich Research Laboratory Säumerstrasse 4 CH-8803 Rüschlikon/Switzerland phone +41-1-724-8698 fax +41-1-724-8578

______________________________________________________________________ _______________________________________________ Forces-protocol mailing list Forces-protocol at ietf.org https://www1.ietf.org/mailman/listinfo/forces-protocol

Re: [Fwd: [Forces-protocol] Presentation of the options forLFB-level multicast]

Note: Messages sent to this list are the opinions of the senders and do not imply endorsement by the IETF.