Re: [Dime] [dime] #45: Why is a validity duration of 0 disallowed?

["TROTTIN, JEAN-JACQUES (JEAN-JACQUES)" <[e-mail]>]

Hi MCruz, Steve

I here  answer Steve and MCruz mails

About Steve, I have not said the reporting "needs" to send a 0% reduction valid
( it is an implementation choice of the reporting node to choose the % values it puts in OLR), I simply says  to not forbid it.

In an overload condition, a behavior example is that the reporting , by its own logic (not standardised) , determines the % reduction  value it send in OLRs, and then will observe the effect of this OLR in the evolution of the received  traffic that  the reacting has throttled. Reporting will then decides to keep, increase or decrease the OLR %value .When we come to the end of overload, the reporting still receiving throttled traffic, will consider that it does  not more need to request a % reduction.

-          One way is to send an OLR with validity period to 0 to indicate the end of the overload condition, but it does this without having yet received unthrottled  traffic that in some case may nevertheless  request to immediately send again an OLR requesting reduction, meaning that reporting, having terminated  an overload condition, will  immediately restart another one. This a way to proceed as Steve described , I do not say it does not work

-          The way I describe is that  reporting  sends a 0% value to see the effect of this new value on the traffic it will receive (that will be unthrottled), before concluding  to the end of the overload , and if nevertheless the received unthrottled  traffic still require to send  OLRs,  reporting will not conclude to the end of overload condition  and again generate OLRs with a non zero % value.

On the reacting, if it receives a 0% value, it simply does nothing,

I remind  I agree with the proposal  that the end of the Overlaod condition is determined by the Value 0 of the Validity period , or the end of the  expiry timer, but is not linked to the 0%  traffic reduction

Developers, have various possibilities in the  way   to manage the  overload condition  and the OLRS they send, I think that DOIC should leave flexibility.

Best regards

JJacques


De : DiME [mailto:[e-mail]] De la part de Steve Donovan
Envoyé : jeudi 13 février 2014 13:16
À : [e-mail]
Objet : Re: [Dime] [dime] #45: Why is a validity duration of 0 disallowed?

JJ,

Why does the reacting node need to send a reduction percentage of zero to determine if it is stable?  Can't the reacting node just do its normal checking for overload after sending the validity period of zero?  If it finds the need for reduction of traffic, it can just send a new overload report.

Steve
On 2/13/14 4:43 AM, TROTTIN, JEAN-JACQUES (JEAN-JACQUES) wrote:
Hi MCruz

AS  Ben proposed, and I agree,   value of zero (0) for the Validity period indicates that an existing overload condition has ended and that the reporting node is in a stable condition.
An important word is "stable".

When the traffic peak on client side having created the overload decreases,   the reporting node progressively diminishes the % traffic reduction in OLRs it sends  towards clients taking care to minimize the possible oscillations. At a certain time ,  server will consider that it can indicate no traffic reduction to clients. Then is it stable?  there may be different implementation dependent  ways for the server to decide the situation is stable;  one is to send 0% in OLRs and wait a certain number of seconds (not 24 hours) to check if situation is stable  and then put the validity timer to 0 (or leaves the expiry timer  expire). I do not see why to forbid this  way to test the stable condition.

So the end of overload condition, as Ben proposed,  can remain ONLY based on Validity  duration value 0 (or timer expiry), not on the value 0 of the % reduction (so a bit different from Nirav statement, but in line with Ulrich comment) . The value 0% of the traffic reduction is not forbidden as a possible  way to check the stability condition .

Best regards

JJacques


De : DiME [mailto:[e-mail]] De la part de Maria Cruz Bartolome
Envoyé : jeudi 13 février 2014 10:56
À : [e-mail]<mailto:[e-mail]> list
Objet : Re: [Dime] [dime] #45: Why is a validity duration of 0 disallowed?

Hello all,

I think proposal to use just Validity-Duration=0 to end overload mainly has the intention to simplify and avoid the checks you just listed below.
If we allow both, it means that the case Ulrich mentioned is valid, and even with 0% reduction OLR info cannot be deleted until Validity time expires, even we could receive a new OLR in sequence. Even, the reporting needs still to keep Validity timer on for this OLR.
I think this does not provide any added value but simply makes things a bit more complex. What could be the reason to keep 0% reduction but a Validity of a few hours (e.g.)?

Best regards
/MCruz


From: DiME [mailto:[e-mail]] On Behalf Of TROTTIN, JEAN-JACQUES (JEAN-JACQUES)
Sent: jueves, 13 de febrero de 2014 10:24
To: [e-mail]<mailto:[e-mail]> list
Subject: Re: [Dime] [dime] #45: Why is a validity duration of 0 disallowed?

I am OK with Ulrich

JJacques


De : Wiehe, Ulrich (NSN - DE/Munich) [mailto:[e-mail]]
Envoyé : jeudi 13 février 2014 09:56
À : ext Nirav Salot (nsalot); TROTTIN, JEAN-JACQUES (JEAN-JACQUES); [e-mail]<mailto:[e-mail]> list
Objet : RE: [Dime] [dime] #45: Why is a validity duration of 0 disallowed?

I also agree with the principle.

One minor clarification:
0%-reduction with non-zero validity period is valid but validity period cannot be ignored: as long as not expired the sequence number needs to be stored for future checking; once expired, sequence number must not be used for future checking.

Ulrich

From: DiME [mailto:[e-mail]] On Behalf Of ext Nirav Salot (nsalot)
Sent: Thursday, February 13, 2014 4:11 AM
To: TROTTIN, JEAN-JACQUES (JEAN-JACQUES); [e-mail]<mailto:[e-mail]> list
Subject: Re: [Dime] [dime] #45: Why is a validity duration of 0 disallowed?

I also tend to agree with JJ below.
Unless there is a strong reason, no point in forbidding the use of 0% reduction - which can also indicate end of overload condition.

May be we can clarify that 0% reduction and/or 0 validity period indicates end of overload. In my view, both are valid for the use, individually and together. So,

-          0 reduction, non-zero validity period => Valid. The reacting node can ignore the validity period if reduction is 0.

-          Non-zero reduction, 0 validity period => Valid. The reacting node can ignore the reduction if validity period is 0.

-          0 reduction, 0 validity period => Valid as well. Generally, this is what the reporting node will use to indicate the end of overload condition.

Regards,
Nirav.

From: DiME [mailto:[e-mail]] On Behalf Of TROTTIN, JEAN-JACQUES (JEAN-JACQUES)
Sent: Thursday, February 13, 2014 2:18 AM
To: [e-mail]<mailto:[e-mail]> list
Subject: Re: [Dime] [dime] #45: Why is a validity duration of 0 disallowed?

Dear all

On this ticket,  I agree on Ben's  proposal  to use the Validity period of 0 to indicate the end of overload.
But about the value of 0% reduction why to forbid it ?

In the process to  return to normal traffic conditions, the  server still sending OLR eg with  5% reduction    will consider to no request anymore traffic reduction but without being yet sure if  it will be  stable (end of overload condition as Ben reminded means stable  situation without traffic reduction ) , so server  can send 0% reduction OLR  but with a validity duration different from 0.
Otherwise it has to  use 1% throttling to check stability and if traffic with the client is 1000 request per second this means 10 requests throttled per second which should not be throttled.
In addition, we do not need to handle the 0% reduction  as an error case (cf Mcruz mail)

Best regards

JJacques


De : DiME [mailto:[e-mail]] De la part de Wiehe, Ulrich (NSN - DE/Munich)
Envoyé : mercredi 12 février 2014 10:22
À : ext Maria Cruz Bartolome; [e-mail]<mailto:[e-mail]> list
Objet : Re: [Dime] [dime] #45: Why is a validity duration of 0 disallowed?

Thank you for the clarification.
I agree.

From: DiME [mailto:[e-mail]] On Behalf Of ext Maria Cruz Bartolome
Sent: Wednesday, February 12, 2014 10:13 AM
To: [e-mail]<mailto:[e-mail]> list
Subject: Re: [Dime] [dime] #45: Why is a validity duration of 0 disallowed?

Dear Ulrich, all,

The proposal was to use OC-Validity-Duration AVP =0 to indicate end of overload, since this could be generally applied to any algorithm.
Then, Reduction-Percentage = 0 should be considered as invalid.
I found this reasonable.

Best regards
/MCruz


From: Wiehe, Ulrich (NSN - DE/Munich) [mailto:[e-mail]]
Sent: miércoles, 12 de febrero de 2014 9:57
To: Maria Cruz Bartolome; [e-mail]<mailto:[e-mail]> list
Subject: RE: [Dime] [dime] #45: Why is a validity duration of 0 disallowed?

I'm confused,

I thought that people were in favour of allowing 0 to indicate end of overload.

Ulrich

From: DiME [mailto:[e-mail]] On Behalf Of ext Maria Cruz Bartolome
Sent: Wednesday, February 12, 2014 9:44 AM
To: [e-mail]<mailto:[e-mail]> list
Subject: Re: [Dime] [dime] #45: Why is a validity duration of 0 disallowed?

Ben, all,

This comment affects as well clause 5.5.2:

   value == 0

      Indicates explicitly the end of overload condition and the
      reacting node SHOULD NOT apply the traffic abatement algorithm
      procedures anymore for the given reporting node (or realm).

This should be modified to explain this value is not valid and what is the expected behavior (i.e. it is discarded).

Best regards
/MCruz

From: DiME [mailto:[e-mail]] On Behalf Of Maria Cruz Bartolome
Sent: martes, 11 de febrero de 2014 14:54
To: [e-mail]<mailto:[e-mail]> list
Subject: Re: [Dime] [dime] #45: Why is a validity duration of 0 disallowed?

Ben,

This approach sounds reasonable to me.
But I would like to clarify what should be the behavior if a Reduction-Percentage=0 is received. This is an invalid value, then I presume it should be discarded by client.

Regards
/MCruz

From: DiME [mailto:[e-mail]] On Behalf Of Ben Campbell
Sent: martes, 11 de febrero de 2014 0:56
To: Jouni Korhonen
Cc: [e-mail]<mailto:[e-mail]> list; [e-mail]<mailto:[e-mail]>
Subject: Re: [Dime] [dime] #45: Why is a validity duration of 0 disallowed?

So, here's some proposed text. (intentionally ignoring the related discussion about handling invalid values):

Section 4.5, first paragraph, last 2 sentences:

Old:

Validity duration values 0 (i.e., 0 seconds) and above 86400 (i.e., 24 hours) MUST NOT be used. Invalid validity duration values are treated as if the OC-Validity-Duration AVP were not present.

New:

Validity duration values above 86400 MUST NOT be used. Invalid validity duration values are treated as if the OC-Validity-Duration AVP were not present. A value of zero (0) indicates that an existing overload condition has ended and that the reporting node is in a stable condition.

Section 4.7, 2nd paragraph:

Old:

 The value of the Reduction-Percentage AVP is between one (1) and one hundred (100).  Values greater than 100 are interpreted as 100.  The value of 100 means that no traffic is expected, i.e. the reporting node is under a severe load and ceases to process any new messages. The Reduction-Percentage AVP MUST be present in an overload report that uses the default abatement algorithm.

Note that there is no zero (0) value defined for the Reduction-Percentage AVP. A zero value would logically indicate that no overload abatement is requested. Instead, reporting nodes use a OC-Validity-Duration AVP value of zero (0) to indicate the end of an overload condition. [Section 4.5]






On Feb 10, 2014, at 4:12 PM, Jouni Korhonen <[e-mail]<mailto:[e-mail]>> wrote:
Just post it here.



On Feb 10, 2014, at 6:25 PM, Ben Campbell <[e-mail]<mailto:[e-mail]>> wrote:
Okay. Does that mean we should assign the issue to me in the tracker?

On Feb 10, 2014, at 10:06 AM, Jouni Korhonen <[e-mail]<mailto:[e-mail]>> wrote:

Ben,

Propose some text and we can see how it fits in.

- Jouni


On Feb 10, 2014, at 5:53 PM, Ben Campbell <[e-mail]<mailto:[e-mail]>> wrote:

On Feb 10, 2014, at 5:12 AM, Jouni Korhonen <[e-mail]<mailto:[e-mail]>> wrote:

On Feb 7, 2014, at 11:54 PM, dime issue tracker <[e-mail]<mailto:[e-mail]>> wrote:
#45: Why is a validity duration of 0 disallowed?

Section 4.5 disallows a validity duration of zero. Why do we want to
disallow that? It would allow a quick way of ending any existing overload
condition without worrying about the semantics of the abatement algorithm.
(We currently use a reduction percentage of zero to end an overload
condition--but that's specific to the loss algorithm and might not make
sense for all future ones.)

Right. Avoiding two ways of ending overload condition was the reason.
I am OK to have validity duration 0 as an additional method ending the
overload condition based on the reasoning above.

I would go further and make duration 0 the _preferred_ way, for two reasons:

1) It's algorithm independent. (reduction-percentage of 0 is specific to algorithms that use reduction percentage.)

2) Explicit signaling of the end of an overload condition becomes semantically identical to the expiration of an overload condition. This allows a simpler implementation.



_______________________________________________
DiME mailing list
[e-mail]<mailto:[e-mail]>
https://www.ietf.org/mailman/listinfo/dime





_______________________________________________

DiME mailing list

[e-mail]<mailto:[e-mail]>

https://www.ietf.org/mailman/listinfo/dime