Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

RE: stds-80220-eval-criteria: RE: 802.20 Fairness Criteria




Anna,

You have raised a very good point on whether the fairness criteria apply within the same class of traffic or across traffic classes.

My understanding was that the criteria would only apply to the best-effort traffic types (e.g. web browsing). As you pointed out, applying the same criteria for traffics with different data rate requirements may not be appropriate. Moreover, we may need other performance metrics (e.g. frame erasure rates and delay jitter etc.) in order to characterize the performance of applications such as video and VoIP. The same performance metrics can also be used to determine, for example, if a VoIP user is in outage or not.

Therefore, my suggestion at this point in time would be to limit the fairness criteria requirement to the users of the same traffic type. In case of a mix of traffic types, you would need as many normalized throughput CDFs plotted against the fairness criteria as the number of traffic types simulated simultaneously. Normalized throughput would be calculated by dividing the user throughput with the average throughout of that class.

Thanks,
Farooq

-----Original Message-----
From: Lai-King Tee [mailto:a.tee@samsung.com]
Sent: Tuesday, November 04, 2003 6:17 PM
To: 'Khan, Farooq (Farooq)'
Cc: 'Joseph Cleveland'; 'Jin-Weon Chang'; aappaji@sta.samsung.com;
stds-80220-eval-criteria@ieee.org
Subject: RE: stds-80220-eval-criteria: RE: 802.20 Fairness Criteria


Hi Farooq,

Thanks very much for the reply. You are correct that I was referring to the
voice outage, but I couldn't find the definition for data outage. Anyway, I
am also wondering what assumptions were made on the traffic mix scenario for
the fairness criteria that you have proposed. Would all the users be running
the same kind of FTP applications? 

If there are a large number of users running low data rate applications such
as conversational voice, telnet or voice messaging. The requirement for data
rate may be lower than 10 kbps in some of those users, while there could
still be users running applications such as FTP, video conferencing that may
require 384 kbps or higher data rates. In this case, the fairness criteria
that are based on a factor of 9 between the best and the worst user
throughput may not be fair enough.   

Best regards,
Anna.  

-----Original Message-----
From: owner-stds-80220-eval-criteria@majordomo.ieee.org
[mailto:owner-stds-80220-eval-criteria@majordomo.ieee.org] On Behalf Of
Khan, Farooq (Farooq)
Sent: Wednesday, October 29, 2003 9:45 AM
To: 'Lai-King Tee'
Cc: 'Joseph Cleveland'; 'Jin-Weon Chang'; aappaji@sta.samsung.com;
stds-80220-eval-criteria@ieee.org
Subject: stds-80220-eval-criteria: RE: 802.20 Fairness Criteria


Anna,

I think you are referring to voice outage in the EV-DV methodology. The data
outage requirement is different than the voice outage.

I assumed that users in outage are not counted towards system capacity.
Therefore, they do not show up in the CDF.

Farooq

-----Original Message-----
From: Lai-King Tee [mailto:a.tee@samsung.com]
Sent: Wednesday, October 29, 2003 12:07 PM
To: 'Khan, Farooq (Farooq)'
Cc: 'Joseph Cleveland'; 'Jin-Weon Chang'; aappaji@sta.samsung.com;
stds-80220-eval-criteria@ieee.org
Subject: RE: 802.20 Fairness Criteria


Hi Farooq,

Thanks for your reply. The outage criteria in the 1xEV-DV evaluation
methodology document (Appendix C) defined the per-user and system outage, in
terms of probability that the user's short-term FER exceeded the per-user
outage threshold, and the probability that the percentage of users that were
in outage exceeded the system outage threshold respectively. How would you
relate that to the outage constraint of 9.6 kbps, as you have mentioned in
your email?

Table 1 and figure 1 of your contribution showed that there should be no
user (cdf=0.0) with normalized throughput below 0.2, or 10 kbps in the
example in your email. But the paragraph below the table stated that the
percentage of users in outage (i.e., with normalized throughput below 0.2)
shall be below 2%. Should the CDF corresponding to the normalized throughput
of 0.2 be 0.02 instead of 0.0, for consistency?

Best regards,
Anna.

-----Original Message-----
From: Khan, Farooq (Farooq) [mailto:fkhan1@lucent.com] 
Sent: Wednesday, October 29, 2003 2:48 AM
To: 'Lai-King Tee'
Cc: 'Joseph Cleveland'; 'Jin-Weon Chang'; aappaji@sta.samsung.com
Subject: 802.20 Fairness Criteria

Anna,

Thanks for your comments.
You are right, the criteria that I proposed is more stringent than one in
1xEV-DV. I came up with this criteria based on my experience of using DV
criteria. It turns out that the DV criteria is "too relaxed". In fact the DV
criteria was developed to discourage people not to use "Max C/I" type of
schedulers for artificially boosting the system capacity. 

The new criteria still allows up to a factor of 9 difference between the
best and the worse user throughput (and delay). For example, the criteria is
satisfied if the throughputs are uniformly distributed between 10-90Kb/s
with average throughput of 50Kb/s. A normalized throughput of 0.2 means user
average throughput of 10Kb/s (0.2*5Kb/s). 

I decided to use normalized throughput as outage criteria in contrast to an
absolute number (like in DV we have 9.6Kb/s as the outage constraint) in
order to simplify the specifications. Looking at the above example, users
with throughput below 10Kb/s would be in outage.

Please let me know if you have any other questions.

Farooq

-----Original Message-----
From: Lai-King Tee [mailto:a.tee@samsung.com]
Sent: Tuesday, October 28, 2003 7:22 PM
To: 'Khan, Farooq (Farooq)'
Cc: 'Joseph Cleveland'; 'Jin-Weon Chang'; aappaji@sta.samsung.com
Subject: RE: stds-80220-eval-criteria: Evaluation Criteria Conference
Call - Tuesday, 28 October 2003, 6 :00-8:00 PM Eastern


Dear Farooq:

I was on the conference call earlier today. I have taken a closer look at
the fairness criteria, and compared it with the one described in 1xEV-DV
Evaluation methodology document. The fairness criteria in 1xEV-DV had an
offset of 0.2 to the left of the one that you have proposed. Would you mind
to explain the reason for the difference? The current fairness criteria seem
to be more stringent in the way that you have proposed. 

The other question is about the minimum normalized throughput on your
fairness criteria is 0.2, while the outage would happen when the percentage
of user with normalized throughput below 0.2 is greater than or equal to 2%.
Is there any connection between those two numbers, i.e., cdf at 0.2 vs
outage percentage? 

Thanks very much for your time.

Best regards,
Anna.