p2psp team mailing list archive

[Ilshat Shakirov <im.shakirov@xxxxxxxxx>]

Hello!,

I got a new results from experiments with new parameters (2 trusted and one
attacker), and it doesn't match with expected results from
http://p2psp.org/en/p2psp-protocol?cap=indexsu11.xht#x20-160004.11 (formula
11)
But I deduced new formula (only for these params for now (T=2, A=1)):

​
, where n is team size.
​Let me explain it:
It just a formula for expected value for discrete random variable. So let
describe our discrete variable:
We have n peers, 2 trusted peers and one attacker. Size of each node's peer
list is* n - 1*. Number of possible permutations for 2 trusted peers in
attacker's peer list is *N = (n - 1) * (n - 2) / 2.*
The probability for getting only one poisoned chunk in the netowrk is *P(1)
= (n - 2) / N*; (because we're getting only one poisoned chunk when trusted
peer is the first in attacker's peer list, so there is n-2 variants of this
situation).
Next, *P(2) = (n - 3) / N*, P(x) =* (n - x - 1) / N*
The max possible number of poisoned chunks in network with 2 trusted peer
and one attacker is n - 2 (n - attacker - trusted);
So, we must only to compute expected value as sum of *P(x)*x*.

Formula can be easily generalized to different numbers of attackers and
trusted peers, but first I want to know your opinion about it.
What do you think? =)
<https://en.wikipedia.org/wiki/Discrete_random_variable>


2015-03-29 1:43 GMT+05:00 Ilshat Shakirov <im.shakirov@xxxxxxxxx>:

> Hello!,
>
> Sorry for the delay in the response.
> I have just improved the Peersim impl of P2PSP:
> - malicious peer connects to the team after some delay (magic constant :))
> - *CHUNK *message have 4 times more delay than other messages (hello,
> goodbye, etc.)
> - trusted and malicious peers are chosen randomly at the begin of
> simulation
>
> I have performed new experiments with these improvements, it can be
> accessed at the same doc -
> https://docs.google.com/spreadsheets/d/1pc6yb87xJy8gNkWSWvvCvjAjR6WBFdOzCbPvO-zEooU/edit#gid=0
> - it has 2 new tabs - raw2 and result2.
> *raw2 *tab have 2 sets of data - the first one was performed without
> different latencies, and the second one was performed with different
> latencies.
> Could you please evaluate results? What the next steps you want to see
> with peersim impl of p2psp? =) I want to perform experiments with different
> numbers of trusted peers and attackers, in order to confirm theroretical
> analysis given in paper.
>
> Also, I want to store all the logs from each experiment in the same place.
> I want to use dropbox for this (shared dir). What do you think? =)
>
> Currently, Im preparing scripts (bash and bat) to run simulation, so there
> will be possible to perform experiments without setting up IDE.
>
> ps Could you please include me to sim project on github? Or we will work
> with pull requests? =)
>
> Thanks in advance!
>
>
> 2015-03-20 18:48 GMT+06:00 Vicente Gonzalez <
> vicente.gonzalez.ruiz@xxxxxxxxx>:
>
>>
>>
>> 2015-03-20 9:59 GMT+01:00 Ilshat Shakirov <im.shakirov@xxxxxxxxx>:
>>
>> I don't know the procedure you are using to add peers to a team (how the
>>>> peers contact the splitter, receives the list of peers of the splitter and
>>>> send the [hello] message to the peers that are in the list), but I suppose
>>>> that this procedure is similar to one used in a real team.
>>>
>>> Yes, the peers are added to team as it described in DBS of rules (new
>>> peer sends HELLO to splitter, splitter sends list of peers, peer sends
>>> HELLO to rest of team).
>>>
>>
>> OK.
>>
>>
>>> I have new hypothesis - how much time is spent for sending chunks? and
>>> for sending simple messages, like HELLO/GOODBYE. In peersim implementation
>>> there is no difference in times between these two types of message. If it's
>>> able to send 2 or more simple messages while one chunk is sending, then
>>> simulation results will match with expected results.
>>>
>>
>> When a peer is introducing himself to the rest of the team only sends
>> [hello]s. To interlace [hello]s and chunks, the splitter should send first
>> to it a chunk, and this is done after receiving the list of peers from the
>> splitter. This does not happen.
>>
>> In my opinion, a better result (not only for this experiment) could be
>> obtained if you assign a latency to each message that depends on the
>> message's length.
>>
>>
>>> This means that if the peer 34 were the trusted peer, the attacker would
>>>> send only one poisoned chunk before it is reported. For this reason, I
>>>> think that the order of the peers in the lists should have some influence
>>>> in the average number of poisoned chunks.
>>>
>>> There is such results already. I.e. cell D43 in google sheet
>>> <https://docs.google.com/spreadsheets/d/1pc6yb87xJy8gNkWSWvvCvjAjR6WBFdOzCbPvO-zEooU/edit#gid=0>.
>>>
>>>
>>
>> I see :-)
>>
>>
>>>
>>> Also, I can remove the latency for simple messages (like hello and
>>> goodbye), and (I suppose) results will match with expected results.
>>>
>>
>> No. Please, consider first my idea (longer messages will take more time
>> traveling that shorter ones).  And, randomize, if possible, the order (or
>> time) in which peers join the team.
>>
>>
>>>
>>> An accurate P2PSP simulator is very interesting for researching the
>>>> performance of the P2PSP. Any kind of advance/facility in this direction is
>>>> really appreciated :-)
>>>
>>>
>>
>>> Ok, I will begin to write readme file about performing experiments =)
>>>
>>
>> Again, thanks. Your work will be very helpful!!
>>
>> Best,
>> Vi.
>>
>> --
>> Vicente González Ruiz
>> Depto de Informática
>> Escuela Técnica Superior de Ingeniería
>> Universidad de Almería
>>
>> Carretera Sacramento S/N
>> 04120, La Cañada de San Urbano
>> Almería, España
>>
>> e-mail: vruiz@xxxxxx
>> http://www.ual.es/~vruiz
>> tel: +34 950 015711
>> fax: +34 950 015486
>>
>
>