Re: [IPFIX] draft-ietf-ipfix-reducing-redundancy-04.txt: text clarification
Brian Trammell <trammell <at> tik.ee.ethz.ch>
2009-02-12 09:14:08 GMT
Hi, Benoit,
Two minor points:
1. Change "...the use of PR-SCTP with the full reliability..." to
"...the use of PR-SCTP with full reliability..." (no need for the
article). Oops, sorry about that.
2. This proposed change is much, much clearer. However, I'm still not
convinced that
(a) the recommendation against UDP is _really_ dependent on the
loss characteristics of the lower layer; or that
(b) on a second reading, this should really be a recommendation
against UDP, but rather a caveat given that UDP has already been
chosen (after all, if you're using UDP, you're doing NetFlow
replacement, you've presumably already read Section 6.2 of RFC5153 and
know all the caveats, and you're going to use UDP anyway). So I think
the recommendation here would be clearer and more internally
consistent if it restated the caveats in Section 4.2, instead:
So, I'd suggest:
NEW:
The proposed method is most effective using a reliable transport
protocol for the transfer
of the Common Properties. Therefore the use of PR-SCTP with full
reliability or TCP
is recommended for the transmission of IPFIX Messages containing
Common Properties.
Note that this method loses some of its effectiveness if UDP is
used for the transmission
of IPFIX Messages containing Common Properties, since the Common
Properties MUST then,
like Templates, be re-sent at regular intervals as in Section 4.2.
Cheers,
Brian
On Feb 12, 2009, at 9:38 AM, Benoit Claise wrote:
> Brian,
>
> First of all, I would like to have Sven's feedback, as he was the
> one asking for this change.
> Then, see inline.
>> Hi Benoit, all,
>>
>> Hm, that's still kind of confusing. First off, there's no "reliable
>> mode" in PR-SCTP. SCTP is modeless, reliability is per-message. PR-
>> SCTP can provide "fully reliable" transport. But this terminology
>> seems to be overloaded:
>>
>> "If the path from the Exporting Process to the Collecting Process
>> is full reliable"
>>
>> What does this mean, precisely? My assumption is that it means "if
>> the link between the Exporting Process and the Collecting Process
>> has low loss and reordering characteristics..." However, the
>> guidance doesn't actually change depending on the loss
>> characteristics of the lower layer: RR prefers reliable transport
>> for Common Properties, full stop. I'd suggest (as usual...)
>> removing any mention of UDP as recommended:
> The recommendation doesn't change, agreed.
> But the sentence is a warning/clarification against the use of UDP,
> even if the path is fully reliable.
> So my preference is to keep this sentence (slightly modified, see
> below)... however, this is a minor point IMHO.
> What are the others thinking?
>
>
>>
>> NEW:
>> The proposed method is most effective using a reliable transport
>> protocol for the transfer of the Common Properties. Therefore the
>> use of PR-SCTP with the full reliability or TCP is recommended for
>> the transmission of IPFIX Messages containing Common Properties.
> NEW:
> The proposed method is most effective using a reliable transport
> protocol for the transfer
> of the Common Properties. Therefore the use of PR-SCTP with the
> full reliability or TCP
> is recommended for the transmission of IPFIX Messages containing
> Common Properties.
> If the path from the Exporting Process to the Collecting Process
> is fully reliable (i.e., no loss),
> the use of UDP is less effective because the Common Properties
> have to be re-sent regularly.
>
>
> Regards, Benoit.
>>
>> Cheers,
>>
>> Brian
>>
>> On Feb 11, 2009, at 4:40 PM, Benoit Claise wrote:
>>
>>> Dear all,
>>>
>>> We received the following remark from Sven regarding http://www.ietf.org/internet-drafts/draft-ietf-ipfix-reducing-redundancy-04.txt
>>>> 8.1. Transport Protocol Choice
>>>>
>>>> The proposed method is most effective using a reliable transport
>>>> protocol for the transfer of the Common Properties. Therefore the
>>>> use of PR-SCTP with the reliable mode or TCP is recommended.
>>>> However, if the path from the Exporting Process to the Collecting
>>>> Process is not fully reliable, the SCTP or TCP retransmission
>>>> might
>>>> reduce the benefits of this specification. If the path from the
>>>> Exporting Process to the Collecting Process is full reliable,
>>>> the use
>>>> of UDP is less effective because the Common Properties have to
>>>> be re-
>>>> sent regularly.
>>>
>>> Do I understand that section correctly: If the path is reliable,
>>> you should use a reliable transport, and if the path is unreliable
>>> you should use unreliable UDP? Isn't that upside down? If the path
>>> is reliable, you can use UDP _without_ retransmission, no need for
>>> a reliable transport protocol then.
>>> We proposed the following text, removing the confusing sentence.
>>> NEW:
>>>
>>> The proposed method is most effective using a reliable transport
>>> protocol for the transfer of the Common Properties. Therefore the
>>> use of PR-SCTP with the reliable mode or TCP is recommended.
>>> If the path from the Exporting Process to the Collecting Process
>>> is
>>> full reliable, the use of UDP is less effective because the Common
>>> Properties have to be re-sent regularly.
>>>
>>> Any objections?
>>> Pending on nobody objecting in a couple of work days, this
>>> editorial disambiguation will be applied.
>>>
>>> Regards, Benoit.
>>> _______________________________________________
>>> IPFIX mailing list
>>> IPFIX <at> ietf.org
>>> https://www.ietf.org/mailman/listinfo/ipfix
_______________________________________________
IPFIX mailing list
IPFIX <at> ietf.org
https://www.ietf.org/mailman/listinfo/ipfix
RSS Feed