Re: ForCES TLV size

We chose 16 bits for the length because the VAST 
majority of things will fit comfortably within that.
(and because a 16 bit type and a 16 bit length totals to a nice 32 bit set)
We could just change the TLV/ILV to use 32 bit 
lengths.  But that seemed to be quite excessive.
Also, sending a single objet that large, in a 
single block, does strike me as asking for something to go wrong.

So we have proposed a mechanism to break up 
over-large things in a special case.

Or we could do nothing at all.  The problem is 
likely rare enough we could survive without a fix for now.

Up to the working group.

Yours,
Joel

At 12:56 AM 8/17/2007, Wang,Weiming wrote:
>Jamal，
>
>I think  you catched a very important point, 
>i.e., many TLVs we'v defined may also face the 
>insufficienct length problem with 16bits space, 
>like the Redirect TLV, LFBselect TLV, as well as the Oper TLV.
>
>Whereas, I don't think 32 bits length is still insufficient.
>
>I just think that if we can just define a 
>universal TLV structure like a ILV we'v defined? 
>In this way, the change to the text is also easier.
>
>I hope more people can join in the discussion.
>
>thanks,
>Weiming
>
>
>----- Original Message -----
>From: "Jamal Hadi Salim" <[email protected]>
>
>Hi Weiming,
>
>There is a small problem:
>The LFB select TLV "L" part is 16 bits. So the underlying data length
>cannot exceed 16 bits. We can change that too to be an ILV, but then i
>think at some point when 32 bit is no longer sufficient (infiniband has
>64K MTU for example) then this wont solve it either.
>
>cheers,
>jamal