SCE loses Quota Below Threshold events in cutoff redundancy mode

[k.vlachoulis]

Hi all,

We are implementing a inline cascade topology with 2 SCE 2020 devices and the action on the SCE failure is cutoff and we are experiencing delay in "Quota Below Threshold" events.The events consequently generated upon quota depletion (i.e. "Quota Below Threshold" and "Quota Depleted") do appear, but meanwhile the subscriber experiences temporarily loss of service. The most important implication is that the subscriber loses all open downloading sessions. We haven't experienced such an issue while we were working in the "bypass" redundancy mode.

Here follows an excerpt of our logs which indicate the problem while a subscriber has been downloading a file. The logs are generated by the API quota listeners at the lowest level. The quotas that get replenished each time are 50Mb. The quota threshold was set to 20Mb (from the default 10Mb) in an effort to mask off the problem, but unfortunately it persisted. The downloading rate wasn't high enough to justify a rapid replenishment of quotas.

...

2011-05-06 17:40:26,499 DEBUG [handlers.SceQuotaListener] Handling Quota event for subscriber xxx with quota 16426 and type QUOTA_EVENT_TYPE_STATUS

2011-05-06 17:40:41,491 DEBUG [handlers.SceQuotaListener] Handling Quota event for subscriber xxx with quota 10161 and type QUOTA_EVENT_TYPE_BELOW_THRESHOLD

2011-05-06 17:41:41,759 DEBUG [handlers.SceQuotaListener] Handling Quota event for subscriber xxx with quota 7668 and type QUOTA_EVENT_TYPE_BELOW_THRESHOLD

[... QBT event  ???]

2011-05-06 17:42:41,819 DEBUG [handlers.SceQuotaListener] Handling Quota event for subscriber xxx with quota -910 and type QUOTA_EVENT_TYPE_BELOW_THRESHOLD

2011-05-06 17:42:41,823 DEBUG [handlers.SceQuotaListener] Handling Quota event for subscriber xxx with quota -910 and type QUOTA_EVENT_TYPE_DEPLETED ...

We are of the impression that the quota below threshold or depletion conditions are detected after a small delay of 2-3 seconds. Is it possible that this detection gets delayed somehow under certain conditions?

Thank you in advance

[Shelley Bhalla]

Rapid use of quota does explain this behavior but seems like you are already aware of this situation.

Just to be sure, The Threshold should be set by evaluating the volume that the fastest subscriber in the system can consume in 30 seconds period. In your policy this would be both upstream + downstream the fastest subscriber can get (since quota is calculated in both directions). So the volume which the subscriber can consume with such a rate for 30 second period should be a guideline for setting the threshold value (threshold should be set significantly higher than this value).

The QM uses a Sliding Window Model for measuring subscriber consumption -

http://www.cisco.com/en/US/docs/cable/serv_exch/serv_control/broadband_app/rel36x/qm_sol/01_overview.html#wp1053153

The Remaining Quota interval is configured in the SCA-BB GUI (at the RDRs section). What have you configured it as?

Can you increase the quota replenishment limit to 100megs from 50 megs and leave the below threshold limit to 20 megs.

------

Also not sure if you already have this but this link and the paragraph below may help:

http://www.cisco.com/en/US/docs/cable/serv_exch/serv_control/broadband_app/rel355/qm_sol/02_scenarios.html#wp1053156

Maximizing Quota Accuracy

One of the most important aspects of the quota manager is accuracy of the quota levels for any subscriber. When you provision quota using an external server, a trade-off exists between quota accuracy and the number of network messages.

To maximize accuracy, configure the rate of the periodic remaining quota indication to a high value, and configure the size of the quota dosage to a small value. A configuration causes performance degradation due to the high number of messages being generated in the network.

Quota inaccuracies may occur during the changeover from one aggregation period to the next, or due to SCE fail-over. The level of inaccuracy depends on the configuration of the following parameters:

•Rate of the periodic remaining quota indications

•Quota dosage value

During an aggregation period changeover, the following occurs until the first quota indication is received in the new aggregation period:

•Any quota consumed by the subscriber is subtracted from the previous aggregation period.

•The quota dosage value limits the size of any quota error.

•The interval between the remaining quota indications limits the length of time during which consumed quota is subtracted from the previous aggregation period.

In cases of SCE fail-over, the following occurs between the last quota indication in the failed SCE and the first quota indication in the new, active SCE:

•Any quota consumed by the subscriber is not removed from the subscriber buckets.

•The quota dosage value limits the size of any quota error.

•The length of time during which quota is consumed is limited by the interval between the remaining quota indications.

In all cases of inaccuracy, the quota remaining is calculated in favor of the subscriber. The only exception is if the aggregation period changeover occurs when the subscriber quota is already breached.

Shelley.

[k.vlachoulis]

The Remaining Quota interval is configured in the SCA-BB GUI to 20MB from 10Mb that is the default value.

Do you know what is the size of the window that QM uses in the sliding window model for measuring subscriber consumption?

Up to now we experienced something like 3 second delay to detect QBT or QDEPL in bypass mode. While in cutoff we saw SOMETIMES much more (maybe 30 second-window applied in that case). Why the redundancy mode has different results with quota detection window, The issue is that if we follow the above suggestion to cater for 30 second-traffic, we would need to set threshold to say 60Mb (for 2Mbytes/sec rate) which is too high, because we would also need to increase the quota chunk we download to DPI each time.

Thanks

Kanellos

[Shelley Bhalla]

when you are talking about cutoff and bypass, what are you refering to?

Slice period is defined by the "slice_period" parameter in the quota profile.

http://www.cisco.com/en/US/docs/cable/serv_exch/serv_control/broadband_app/rel36x/qm_sol/03_configuring.html

You can actually default back to the standard quota profile by removing the "slice-period" parameter from the user profile. Try this and let me know.

Shelley.

[k.vlachoulis]

We have configured the two SCEs 2020 in inline cascade and when a failure occurs the action is to cutoff the traffic.

What do you mean by removing the slide period parameter? Do you mean to set the value to 0 or delete this parameter?

Thanks

[Shelley Bhalla · Accepted solution]

Yes pretty much delete that parameter.

Shelley.