Poisson distribution in Mahimahi traces

[Chandan Bothra]

Hello Francis,

Hope you are doing well. Thanks for the great work in Pantheon and making it open source for the community. 

I had a doubt about the generation of traces with Poisson distribution of packet delivery opportunities. 

For a given Poisson distribution with rate L, I was sampling the inter arrival time between two packets using exponential distribution with mean (1/L) and using this for each packet in the mahimahi trace. Is this the right way or am I missing something?

Looking forward to hearing from you.

Thanks,

Chandan

[Francis Y. Yan]

Hi Chandan — Sorry for the late response. Yes, that’s the right way.

Best,

Francis

--
You received this message because you are subscribed to the Google Groups "Pantheon" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pantheon-stanf...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/pantheon-stanford/4a36e970-08b3-4762-8715-33fca5123d2an%40googlegroups.com.

[Chandan Bothra]

Hello Francis,

Thanks for your reply. It helps a lot 😅

Best,

Chandan

[Chandan Bothra]

Hello Francis,

Hope you are doing well. 

I was trying to use Poisson distribution for emulating cellular links, however we are still a bit off our expected bandwidth distribution. Out of curiosity, does Pantheon consider the inter arrival time to be doubly stochastic, i.e., use a normal distribution for rate L with some standard deviation, as mentioned in the paper: https://www.usenix.org/system/files/conference/nsdi13/nsdi13-final113.pdf

Thanks,

Chandan

[Francis Y. Yan]

Hi Chandan,

I did not use any doubly stochastic process. Did you generate the Poisson trace in a similar way as follows (in Python)?

import random

timestamp = 0
average_packets_per_second = 10

while timestamp <= 60000:  # 1 minute long
    timestamp += random.expovariate(average_packets_per_second)
    print(int(timestamp))

Best,

Francis

[Chandan Bothra]

I was trying something similar. Thanks, I'll check this out. 

Best,

Chandan