Re: [ippm] Some though on draft-mirsky-ippm-twamp-light-yang-07

[Henrik Nydell <[e-mail]>]

Some comments from the "field" as Accedian has several hundred thousand
TWAMP sessions running (continously) at numerous Tier one mobile/fixed
operators globally.

On Tue, Mar 21, 2017 at 11:52 AM, Wei Luo S <[e-mail]> wrote:

> Hi Greg,
>
>
>
> Thanks a lot for your response. Please see my reply inline tagged [WEI>>].
>
>
>
> Regards,
>
> Wei Luo
>
>
>
> *From:* Greg Mirsky [mailto:[e-mail]]
> *Sent:* Tuesday, March 21, 2017 1:16 AM
> *To:* Wei Luo S <[e-mail]>
> *Cc:* [e-mail]; [e-mail]
> *Subject:* Re: Some though on draft-mirsky-ippm-twamp-light-yang-07
>
>
>
> Hi Wei Luo,
>
> many thanks for your thorough review and the most helpful comments to the
> TWAMP Light(Test) model. Please find my answers, notes in-line tagged GIM>>.
>
>
>
> Regards,
>
> Greg
>
>
>
> On Sat, Mar 18, 2017 at 4:27 AM, Wei Luo S <[e-mail]> wrote:
>
> Hi Greg & Adrian,
>
>
>
> This is Wei Luo from Ericsson. I work on TWAMP light area in Ericsson. The
> current TWAMP Light YANG model is well defined. Thanks for your great job.
>
> But by working closely with our customers, we got some new user cases on
> TWAMP light. I believe these user cases are valuable and popular enough to
> be modeled in TWAMP Light YANG. I hope I can be a contributor  and co-work
> with you move this draft forward.
>
> I  drafted a new version of the TWAMP light YANG model based on version
> [e-mail]. Could you please comments on it? Any
> discussion is welcome.
>
> The draft yang model and tree is attached. To make you find the updates
> quickly, I highlighted all the updates in file
> ietf-twamp-light-weiluo.pdf.
>
>
>
> The following are the list of main updates:
>
> *1. Add a new typedef: percent. This is a new type defined for packet loss
> ratio.*
>
> Consideration:
>
> 1). From the customer perspective, packet loss ratio is a more meaningful
> data. In most of the time, the absolute number is meaningless to user,
> especially they do the TWAMP test continuously. They are more care about
> the ratio than the absolute number. So adding it makes this model more
> friendly to customer;
>
> 2). From the service layer assurance(SLA) perspective, the packet loss
> ratio is a major measures. So with adding packet loss ratio in model, the
> TWAMP can work in SLA framework more smoothly.
>
> 3). It seems some similar protocol’s YANG model has the same definition,
> e.g. ‘Service OAM Performance Monitoring YANG Module’,
> https://www.mef.net/Assets/Technical_Specifications/PDF/MEF_39.pdf.
>
> Agreed packet loss is important, however another important loss metric is
loss burst size (max/min) and number of loss bursts. A loss burst of 10
consecutive TWAMP-test packets can be deemed more serious than 10 lost
packets spread evenly over the report interval.

> GIM>> Indeed, packet loss more often expressed as packet loss ratio rather
> than as the absolute number. It would be most helpful to hear from network
> operators if they see introduction of Packet Loss Ratio into the TWAMP
> model helpful.
>
> *2. Add a new typedef: state-mode. It defines a common type for
> stateful/stateless reflector. This type will be used in both sender session
> and reflector session.*
>
> Consideration:
>
> If the reflector is stateful, the TWAMP light can measure more items, e.g.
> one way packet loss. So for sender, the stats calculation and show is
> different. When the reflector is stateless, it doesn’t need to calculate
> the one way packet loss. The one way packet loss is invalid and shouldn’t
> be presented to customer. When the reflector is stateless, the sender needs
> to calculate the one way packet loss. And the data should be present to
> customer. So this is used as a ‘when’ condition in the model’s RO tree.
>
> GIM>> Yes, if Session-Sender is aware of the mode corresponding
> Session-Reflector operates, the sender may avoid calculation of some
> performance metrics, e.g., one-way packet loss. On the other hand, the
> orchestrator is aware of the state-mode and should be capable to properly
> use metrics reported by the Session-Sender.
>
> [WEI>>] Yes, the orchestrator could know that. But from the model side,
> this is not correct.  The model should represent the right behavior and
> shouldn’t do assumption on orchestrator.
>
I agree the model should describe both one-way loss metrics and roundtrip
loss metrics, and the sender should be able to use either mode when
calculating, potentially also populating the roundtrip delay values with
proper t1-t0 + t3-t2 values, as well as reporting the t2-t1 values that
would indicate buffer load/CPU load in the TWAMP responders processing time.

> *3. Add a new typedef: send-mode. This is a new type for sender session.
> It makes the sender session can send packet continuously and monitor the
> network all the time.*
>
> Consideration:
>
> The user case is that: the user runs TWAMP light sessions to watch links
> quality continuously. The session number could be very big. These TWAMP
> sessions are managed by SLA framework or similar. SLA retrieves the stats
> from TWAMP periodically, e.g. 15mins. In other words, all the performance
> metrics are calculated based on the packets sent/received within 15mins.
> This makes the calculation become possible. With the periodical stats data,
> the Network Management software can do further actions if some abnormal
> stats observed.  This is a more general user case in customer site. While
> the non-continuous TWAMP sender session is generally used for debugging
> purpose on a link.
>
> GIM>> I think that support of continuous measurement is in LMAP domain,
> not for TWAMP Test data model. To conduct continuous measurement he LMAP
> Controller, in my opinion, programs the Measurement Agent to perform TWAMP
> Test session with certain set of parameters and repeat it without any
> interval (interval = 0).
>

Many operators use TWAMP in continous mode, not only with Accedian test
points and report at fixed intervals, typically ranging from 5s to 5 or 15
minutes, with 1-minute being the most popular granularity currently. The
advantage is that the result calculation can be handled separately from the
TWAMP-test sending/recieving, so that there is no parallelism required to
monitor 24/7. If a start-stop-based methodology is used, the sender needs
to start up the new test session even before the previous one has ended,
since the previous session needs to wait X seconds (or at least Y 100s of
milliseconds) before it stops waiting for packets to come back. And this
new session needs to have a different signature in order for the sender to
discern which packets belong to the previous interval and which belong to
the current.

In a continous test-model, the sender can just simply record the sequence
number of the last packet transmitted in the interval to be reported, wait
for it to come back, or a MAXTIME, then report that result, while
continuing to transmit for the next interval.

If the "interval==0" parameter is intended to be used for continous type
tests, then what parameter should indicate to the sender at what intervals
to produce results?

> *4. Add a new group: packet-loss-statistics. It grouping two packet loss
> statistics: loss-count and loss-ratio. This group will be used in RO stats
> tree.*
>
> GIM>> I'd like to continue discussion.
>
> [WEI>>] OK.
>
> *5. Move leaf dscp out from grouping session-light-parameters. The leaf
> dscp is only valid when the dscp-handling-mode is use-configured-value. A
> when condition shall be added to it. So it can’t be in this group.*
>
> GIM>> I'm concerned that then the model will not be able to support
> concurrent TWAMP Test sessions between the same pair of Test Points (IP
> address+port number) at different CoS markings.
>
> [WEI>>] Actually, I have concern on using five tuple(IP address+port
> number+dscp) to identify a TWAMP test session. The DSCP is not a constant
> value in packet. It could be modified by the routers in the path. For
> example, the sender has two sessions: session A’s five tuple is:
> Sip=1.1.1.1, Dip=2.2.2.2, Sport=50000, Dport=50001, DSCP=cs2. Session B’s
> five tuple is: Sip=1.1.1.1, Dip=2.2.2.2, Sport=50000, Dport=50001,
> DSCP=cs3. The only difference between session A and session B is DSCP. If
> the test packet’s DSCP of session B is modified to cs2 by a router in the
> path. The five tuples are exactly the same for reflector. It can’t
> differentiate which packet is from session A, which packet is from session
> B. It could mess the reflector’s session sequence number. And also, the
> sender will be messed because the received reply packet’s five tuple are
> exactly the same.
>
> So I think it’s more reasonable to use four tuple to identify a session.
>

Yes, this would be appreciated by users. Changes in DSCP is a reasonably
common network error that users can detect with continous TWAMP monitoring,
thus it is good to not include the DSCP value as part of the "session
identifiier" but instead use 4-tuple with UDP source port to identify
several parallel flows between the same sender and responder.

> *6. Add leaf 'session-packet-send-mode' to
> /twamp-light/twamp-light-session-sender/test-session*. This leaf specifies
> the sender session's packet send mode: continuous or non-continuous.*
>
> GIM>> As discussed in #3, I think that it is already part of LMAP YANG
> model.
>
> *7. Add leaf 'reflector-light-mode-state' to
> /twamp-light/twamp-light-session-sender/test-session*. This leaf indicates
> the the reflector's mode: stateful or stateless. If the reflector's mode is
> stateful. Two one way packet loss statistics can be got:
> one-way-packet-loss-far-end, one-way-packet-loss-near-end.*
>
> Consideration:
>
> Only valid data should be presented to user. Otherwise it could misleading
> user in some cases.
>
> GIM>> A in response to #2.
>
> *8. Modify leaf
> /twamp-light/twamp-light-session-sender/test-session*/number-of-packets.
> Add a 'when' condition to this leaf. When send-mode is 'continuous', the
> leaf number-of-packets is meaningless. So add a 'when' condition to limit
> it.  Besides, added a default value ‘10’ to it. When the send-mode is
> 'non-continuous', the session can't work with an empty number-of-packets.*
>
> GIM>> As I've noted in #3. Will add default.
>
> *9. Add leaf time out to
> /twamp-light/twamp-light-session-sender/test-session*. A timeout mechanism
> is needed when the sender session can't get all the reply packets for a
> long time.*
>
> GIM>> Thank you, will add in the next update.
>
> *10. Modify leaf
> /twamp-light/twamp-light-session-sender/test-session*/interval. Change the
> units from ‘microseconds’ to ‘milliseconds’. Add a default value 1000. *
>
> Consideration:
>
>     1). The aim of TWAMP is to measure network quality, but not fast
> failure detection. So a millisecond packet interval is enough.
>
>     2). Interval is a necessary parameter for a session. A sender session
> can't work with an empty packet send interval. So added a default value to
> it.
>
> GIM>> Thank you. We've made units of interval microseconds in the last
> update already. I think that changing to milliseconds may be too
> restrictive, limit use cases for TWAMP Test. Will add default value with
> the next update.
>
> [WEI>>] Sorry, I do not see the reason. Are there any user cases to use
> microseconds?
>
> *11. Add leaf 'dscp' to
> /twamp-light/twamp-light-session-sender/test-session*. This is the leaf
> moved out from grouping session-light-parameters.*
>
> GIM>> As noted in response #5, the change may limit ability to run
> concurrent TWAMP Test sessions per CoS. I consider that to be valuable mode
> but would like to hear from network operators if that is indeed useful
> information.
>

See my comment above. I argue that it is useful to keep track of changing
DSCP values, and treating DSCP as a metric of the TWAMP Session just like
loss and delay

> *12. Move leaves 'ref-wait', 'reflector-light-mode-state' and
> 'dscp-handling-mode' from /twamp-light/twamp-light-session-reflector to
> /twamp-light/twamp-light-session-reflector/test-session*. These three
> attributes should be session specific. Different session could have
> different values. They are not common attributes.*
>
> GIM>> Agree, will make it in the next update.
>
> *13. Add leaf 'dscp' to
> /twamp-light/twamp-light-session-reflector/test-session*. This is the leaf
> moved out from grouping session-light-parameters. Besides the movement,
> added a 'when' condition to the leaf 'dscp'. This leaf is only valid when
> the dscp-handling-mode is 'use-configured-value'.*
>
> GIM>> As response to #5.
>
> *14. Modify leaf
> /twamp-light-state/twamp-light-session-sender-state/test-session-state*/current-stats/number-of-packets.
> Add a 'when' condition to this leaf. When send-mode is 'continuous', the
> leaf number-of-packets is meaningless.*
>
> GIM>> Similar to #3.
>
> *15. Modify leaf
> /twamp-light-state/twamp-light-session-sender-state/test-session-state*/current-stats/interval.
> Change the units from microseconds to milliseconds.*
>
> GIM>> I think that microseconds is reasonable.
>
> *16. Add leaves 'two-way-packet-loss', 'one-way-packet-loss-far-end' and
> 'one-way-packet-loss-near-end' to
> /twamp-light-state/twamp-light-session-sender-state/test-session-state*/current-stats/.
> These are the new statistics for stateful reflector.*
>
> GIM>> Thank you, will be coming in the next update.
>
> *17. Remove leaf loss-packet in
> /twamp-light-state/twamp-light-session-sender-state/test-session-state*/current-stats.
> The loss packeted is replaced with 'two-way-packet-loss' stated above.*
>
> GIM>> Agree.
>
> *18. Modify leaf to
> /twamp-light-state/twamp-light-session-sender-state/test-session-state*/history-stats*/interval.
> Change the units from microseconds to milliseconds.*
>
> GIM>> I think that will limit applicability of TWAMP Test.
>
> *19. Add leaves 'two-way-packet-loss', 'one-way-packet-loss-far-end' and
> 'one-way-packet-loss-near-end' to
> /twamp-light-state/twamp-light-session-sender-state/test-session-state*/history-stats*/.
> These are the new statistics for stateful reflector.*
>
> GIM>> Agree.
>
>
>
> Thanks,
>
> Wei Luo
>
>
>
> _______________________________________________
> ippm mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/ippm
>
>


-- 


[image: Accedian.com]

Henrik Nydell

Sr Manager Global Strategy & Solutions

Cell

Email

Skype

[phone]

[e-mail] <[e-mail]>

h <http://linkedin.com/in/maekowalk>nydell


<http://accedian.com/> <http://blog.accedian.com/>
<https://www.linkedin.com/company/accedian-networks>
<https://twitter.com/Accedian>   <https://www.facebook.com/accedian>
<http://www.youtube.com/user/accedian>

-- 


Avis de confidentialité

Les informations contenues dans le présent message et dans toute pièce qui 
lui est jointe sont confidentielles et peuvent être protégées par le secret 
professionnel. Ces informations sont à l’usage exclusif de son ou de ses 
destinataires. Si vous recevez ce message par erreur, veuillez s’il vous 
plait communiquer immédiatement avec l’expéditeur et en détruire tout 
exemplaire. De plus, il vous est strictement interdit de le divulguer, de 
le distribuer ou de le reproduire sans l’autorisation de l’expéditeur. 
Merci.

Confidentiality notice

This e-mail message and any attachment hereto contain confidential 
information which may be privileged and which is intended for the exclusive 
use of its addressee(s). If you receive this message in error, please 
inform sender immediately and destroy any copy thereof. Furthermore, any 
disclosure, distribution or copying of this message and/or any attachment 
hereto without the consent of the sender is strictly prohibited. Thank you.