benmo009 / multiple-access-queue Goto Github PK

After the last WebEx Meeting, we realized that the age of user 2 was not growing despite having more arrivals.
We think this is because the slot duration assigned to each user was too short for both user 1 and 2, and the consequence of that is the server now is biasing one user over another.

This model is less inefficient compared to the previous issue, because the server right now it is processing all packets from one user then process all packets from another user.

This problem is what needed to be fixed.

We expect by widening the slot duration by a scalar for each user, the age of the higher arrival rate user should grow. Right now we have the slot duration such that it is smaller than the service rate, we are looking to have slot duration of each user to be greater than the service rate.

We will see how it would go from there.

After we scaled up the timeslot that serve each user, we realized that the server tends to prioritize all its time to process the one with greater arrival rate user first and then go back and process the lower arrival rate.

This is not what we wanted, we want that server to finish their tasks within the given time (i.e. 1800s) for BOTH user.

One thing we need to fix right now is to have the server stop working once it hit tfinal (1800), even if there's any packet remaining, no more processing.

Another thing would be to have the code to no longer increase age (or consider user idling) once ALL packets were finished process to show a better age accuracy for the user. (right now we made it such that the user that finished their packets early need to wait until the other user to finish and this is not what we wanted)

Then another thing would be to set in the code that prevent prioritize from happening between a low and high arrival rate user, I believe in doing the age would be much better for the high arrival rate.

FD:

• Adding a weighting factor on the servicing rate impact the performance of the model dramatically here. This resource allocation has more impact than the case with TDMA it seems like.

TD:

• We expect an 'X' shape when both users transmit about the same rate, but if the transmit rate begins to differ greatly, the model will be in favor for the one user with more transmission.

Where do we go from here?

• Clean up the code for FD?

We are wondering:

• A distribution of split factor b vs. changing lambda arrival rate within the appropriate constraint for while keeping the same
• The meaning of the intersection point (where both user's average age is the same), is this good or bad?
• We think that this give us a rough idea of how much resource we needed to assign that would allow both user to perform most of their packets in the given time. If we steer resource allocation more left or right, then their job completion rate would not be the optimal.

May be we need plots for:
Is there a range of optimal split factor where <both> user can serve at least X(percentage > 1) packets/jobs. (favoring unequally)
Is there a range of optimal split factor where <both> user can serve 50% of their packets/jobs (favoring equally)
`Is there a range of optimal split factor where user able to finish their task in the given time?

Then we can say:
Does allocating resources such that both user has the same average age actually is good for both user?

The splitting factor b affects these resources:

1. The split factor of time duration is changed in TDMA
2. The split factor of serve duration is changed in FDMA

To-do:

1. Add count number of packets got through
2. Delay plot is missing included back!
3. Block Diagram for FD
4. TD:
   4.1 Find a range of b such that at least X(percentage > 1) packets/jobs was served
   4.2 Find a range of b such that at least 50% packets/jobs was done

• Questions

So last time when we plot # of packets served / total packets -> that is the utilization factor or the throughput.

For FD:
I think the next step is to superimpose theoretical plot on top of the plot to see how they match.

For TD:
Function is little tricky.

In the beginning, we have corrected that only process user 1 packet if server is at that time it is designed to serve user 1, and vice versa for user 2.

However, we did not consider the time at which packet finish processing. (i.e. If server designed for user 1 started to process user 1's packet, packet can ONLY exit the process if the current time is at a slot with server designated for server 1)