See a few comments (marked GF)
from the perspective of other transport RFCs, in case this helps
you find text...
-------- Forwarded Message --------
-------- Forwarded Message --------
| Subject: | Re: [tcpm] Genart last call review of draft-ietf-tcpm-rto-consider-14 |
|---|---|
| Date: | Thu, 18 Jun 2020 11:00:15 +0100 |
| From: | Stewart Bryant <stewart.bryant@xxxxxxxxx> |
| To: | Martin Duke <martin.h.duke@xxxxxxxxx> |
| CC: | tcpm <tcpm@xxxxxxxx>, Review Team <gen-art@xxxxxxxx>, Mark Allman <mallman@xxxxxxxx>, Last Call <last-call@xxxxxxxx>, Stewart Bryant <stewart.bryant@xxxxxxxxx>, tom petch <daedulus@xxxxxxxxxxxxx>, draft-ietf-tcpm-rto-consider.all@xxxxxxxx |
On 17 Jun 2020, at 18:20, Martin Duke <martin.h.duke@xxxxxxxxx> wrote:
Hi Stewart,
If there are no further objections, I'm going to declare consensus.
On Thu, Jun 11, 2020 at 1:45 PM Martin Duke <martin.h.duke@xxxxxxxxx> wrote:
Stewart,
do we need more cycles for this, or is draft-15 sufficient to address your concerns?
On Mon, Jun 8, 2020 at 12:52 PM Mark Allman <mallman@xxxxxxxx> wrote:
Hi Stewart, et.al.!
I just submitted a new version of rto-consider. Please ask the
datatracker for diffs between this and rev -14. The highlights:
- The diffs with the last rev are here: https://tools.ietf.org/rfcdiff?difftype=--hwdiff&url2=draft-ietf-tcpm-rto-consider-15.txt
In the general case, delay across a
network path depends not only on distance, but also a number of
variable components such as the route and the level of buffering in
intermediate devices.
Its is more the contending/conflicting traffic
rather than the buffering, or perhaps the time spent in
queues, but “buffering” is a link a transport colloquial
term.
GF: The word being sought might be "queueing" (I
think that buffering is thought of as memory- and hence max
queue).
Since our wide-area network paths are best
effort, packet loss is a regular occurrence.
No the best effort Internet experiences this.
There ate many well engineered WAN that do not.
What I am not seeing is clearer text that
distinguishes between user traffic and “engineering” traffic
that is used to make the network work, and between the end
to end traffic and traffic within an AS that may be there
for other purposes (high value service also offered by the
provider) and WANs that are well engineered.
Perhaps we could include a clearer disclaimer
regarding the non-best-effort-internet-end-to-end traffic?
You have some text on this down in section 2 but
it is a bit buried.
Perhaps something early on of the form: This
document is specially concerned with end to end behaviour
over the best effort Internet. As noted in section 2 it may
not me applicable to other types of WAN, or to the traffic
used in affecting the operation of the Internet itself.
GF: Actually, I do think a well-engineering
WAN can be in scope of your spec. The two wrods I was
expecting were "controlled environment" or
"pre-provisioned" capacity, these might not see the same
oath properties. A DC is typically regarded in transport
specs as a "controlled environment".
An exception to this rule is if an IETF standardized mechanism
determines that a particular loss is due to a non-congestion
event (e.g., packet corruption).
That is a bit heavy. It should be “a protocol”
there than an IETF standardarized mechanism. The IETF does
not have a monopoly on pre-blessing protocols before they
are deployed.
GF: Unsure myself what is needed - isn't this guidance for
design of protocol mechansims?
- All small comments addressed.
- I think we all agree that this is not a one-size-fits-all
situation. Rather, this document is meant to be a default case.
So, the main action of this rev is to make that point more
clearly. The first paragraph in the intro is new. Also, there
are some more words fleshing out the context more in section 2.
In particular, more emphatically making the point that other
loss detectors are fine for specific cases.
As I note above from a routing and packet transport (as
opposed to the transport layer) perspective I think we should
more clearly recognise at the beginning the fact that this is
for the worst case network, not for well engineered (WAN and
DC) networks and the mechanisms fundamental to the operation
of the network itself.
- The first paragraph in the intro also makes clear we adopt the
loss == congestion model (as that is the conservative default,
not because it is always true).
- I made one other change that wasn't exactly called for, but
seems like an oversight.
Previously guideline (4) said loss MUST be taken as an
indication of congestion and some standard response taken. But,
this guideline has an explicit exception for cases where we know
the loss was caused by some non-congestion event. Guideline (3)
says you MUST backoff. But, it did not have this exception for
cases where we can tell the cause. But, I think based on the
spirit of (4), (3) should also have these words. So, I added
them.
In some cases you cannot tell the cause, but it is more
important to ignore the loss. OAM being a particularly good
example.
Also, I swapped (3) and (4) because it seemed more natural in
re-reading to first think about taking congestion action and
then dealing with backoff. I think the ordering is a small
thing, but folks can yell and I'll put it back if there is
angst.
Please take a look and let me know if this helps things along or
not.
allman
We are getting there, but I would ask that you take the
transport hat off and look again from an infrastructure and
packet transport perspective.
Best regards
Stewart
On 18/06/2020 11:00, Stewart Bryant
wrote:
On 17 Jun 2020, at 18:20, Martin Duke <martin.h.duke@xxxxxxxxx> wrote:
Hi Stewart,
If there are no further objections, I'm going to declare consensus.
On Thu, Jun 11, 2020 at 1:45 PM Martin Duke <martin.h.duke@xxxxxxxxx> wrote:
Stewart,
do we need more cycles for this, or is draft-15 sufficient to address your concerns?
On Mon, Jun 8, 2020 at 12:52 PM Mark Allman <mallman@xxxxxxxx> wrote:
Hi Stewart, et.al.!
I just submitted a new version of rto-consider. Please ask the
datatracker for diffs between this and rev -14. The highlights:
- The diffs with the last rev are here: https://tools.ietf.org/rfcdiff?difftype=--hwdiff&url2=draft-ietf-tcpm-rto-consider-15.txt
In the general case, delay across a network path depends not only on distance, but also a number of variable components such as the route and the level of buffering in intermediate devices.
Its is more the contending/conflicting traffic rather than the buffering, or perhaps the time spent in queues, but “buffering” is a link a transport colloquial term.
Since our wide-area network paths are best effort, packet loss is a regular occurrence.
No the best effort Internet experiences this. There ate many well engineered WAN that do not.
What I am not seeing is clearer text that distinguishes between user traffic and “engineering” traffic that is used to make the network work, and between the end to end traffic and traffic within an AS that may be there for other purposes (high value service also offered by the provider) and WANs that are well engineered.
Perhaps we could include a clearer disclaimer regarding the non-best-effort-internet-end-to-end traffic?
You have some text on this down in section 2 but it is a bit buried.
Perhaps something early on of the form: This document is specially concerned with end to end behaviour over the best effort Internet. As noted in section 2 it may not me applicable to other types of WAN, or to the traffic used in affecting the operation of the Internet itself.
An exception to this rule is if an IETF standardized mechanism determines that a particular loss is due to a non-congestion event (e.g., packet corruption).
That is a bit heavy. It should be “a protocol” there than an IETF standardarized mechanism. The IETF does not have a monopoly on pre-blessing protocols before they are deployed.
- All small comments addressed.
- I think we all agree that this is not a one-size-fits-all
situation. Rather, this document is meant to be a default case.
So, the main action of this rev is to make that point more
clearly. The first paragraph in the intro is new. Also, there
are some more words fleshing out the context more in section 2.
In particular, more emphatically making the point that other
loss detectors are fine for specific cases.
As I note above from a routing and packet transport (as opposed to the transport layer) perspective I think we should more clearly recognise at the beginning the fact that this is for the worst case network, not for well engineered (WAN and DC) networks and the mechanisms fundamental to the operation of the network itself.
- The first paragraph in the intro also makes clear we adopt the
loss == congestion model (as that is the conservative default,
not because it is always true).
- I made one other change that wasn't exactly called for, but
seems like an oversight.
Previously guideline (4) said loss MUST be taken as an
indication of congestion and some standard response taken. But,
this guideline has an explicit exception for cases where we know
the loss was caused by some non-congestion event. Guideline (3)
says you MUST backoff. But, it did not have this exception for
cases where we can tell the cause. But, I think based on the
spirit of (4), (3) should also have these words. So, I added
them.
In some cases you cannot tell the cause, but it is more important to ignore the loss. OAM being a particularly good example.
Also, I swapped (3) and (4) because it seemed more natural in
re-reading to first think about taking congestion action and
then dealing with backoff. I think the ordering is a small
thing, but folks can yell and I'll put it back if there is
angst.
Please take a look and let me know if this helps things along or
not.
allman
We are getting there, but I would ask that you take the transport hat off and look again from an infrastructure and packet transport perspective.
Best regards
Stewart
_______________________________________________ tcpm mailing list tcpm@xxxxxxxx https://www.ietf.org/mailman/listinfo/tcpm
-- last-call mailing list last-call@xxxxxxxx https://www.ietf.org/mailman/listinfo/last-call
