Updates

[Aaditeshwar Seth]

Haven't heard anything since a long while

--
Aaditeshwar Seth
Co-founder, Gram Vaani Community Media

http://gramvaani.org
http://www.linkedin.com/company/gram-vaani-community-media

[SAMEER A PANDIT]

Sir,
I am working on automated test using selenium.There was initial problems
regarding integration with the suite with the firebug extension and
netexport extension (used to collect har files).I was able fix the
problem with minor changes in netexport code.The code has been deployed
from last 3 days on Delhi nodes (2G/3G).I will update you on the
comparisons of webpage statistics by Wednesday.Also the same test will
deployed at all rural locations by tommorow.(already deployed at lamta)

Things to do:
1.As of now the selenium tests only download(fetching various
webpages),so for uploads I am working on preparing a upload test for
services like gmail,mediafire,dropbox,facebook and will be deployed till
end of the week.

2.Comparison of user experiences for rural location and urban locations
for both upload/download behaviours.


--
SAMEER A PANDIT
Mtech 2nd Year
2012MCS2821

[Aaditeshwar Seth]

What's the update on the buffer behavior tests that had to be redone
with UDP?

[rahul nishant]

Hi Sir,

 I have been working on the analysis of the UDP data that we have collected. I will  be plotting the graphs of the delay and the buffer sizes and giving it in three days time.

Thanks & regards,

Rahul Nishant

--
You received this message because you are subscribed to the Google Groups "ruralnet-act4d" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ruralnet-act4d+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

[SAMEER A PANDIT]

Sir,
Please find attached the report on selenium test at Delhi.

Also Sir,will we be having mid-term presentations ? Please let us know
the schedule for the same.

[Aaditeshwar Seth]

You need to do this from a QoE perspective, by measuring factors that
are expected to contribute to the quality of experience. RTT is only one
factor. You should look at page load time and waiting time for web
browsing, RTT and jitter for VoIP, buffer underruns for video, etc. Take
some cues from these papers:

http://www.research.att.com/techdocs/TD_101214.pdf
http://www.bth.se/fou/forskinfo.nsf/alfs/349bbe221ba80e52c1257a60005503fc/$file/Waiting%20times.pdf

The HAR output will give you access to all these metrics. You can then
take QoE recommendations from other papers to come up with rankings for
different providers. Getting actual user feedback by yourself will be
much harder I think.

[Aaditeshwar Seth]

Also whatever happened to some experiments on CDN. Please go back to my
earlier emails and send a detailed list of next steps.

[Rahul Nishant]

Hi Sir,
Please find attached the updated report on Buffer sizes after the tests
were redone with higher rates.
- report contains the analysis on Buffer sizes and drainsizes.
- The analysis of UDP throughput and inter-packet arrical times needs 2
more days. The scripts for data extraction are taking some time.
- In the downlink direction, the data points obtained are less and nil
for providers like Idea. This is because of private IPs getting alloted
to the measurement nodes. We are trying to find a work around

Thanks & regards,
Rahul Nishant

On 18.02.2014 11:17, rahul nishant wrote:
> Hi Sir,
>  I have been working on the analysis of the UDP data that we have
> collected. I will  be plotting the graphs of the delay and the buffer
> sizes and giving it in three days time.
>
> Thanks & regards,
> Rahul Nishant
>
> On Mon, Feb 17, 2014 at 10:16 PM, Aaditeshwar Seth
> <as...@cse.iitd.ernet.in> wrote:
>
>> What's the update on the buffer behavior tests that had to be redone
>> with UDP?
>>
>> On 17-02-2014 21:36, SAMEER A PANDIT wrote:
>> On 17.02.2014 20:20, Aaditeshwar Seth wrote:
>> Haven't heard anything since a long while
>>
>> --
>> Aaditeshwar Seth
>> Co-founder, Gram Vaani Community Media
>>

>> http://gramvaani.org [1]
>> http://www.linkedin.com/company/gram-vaani-community-media [2]

>>
>> Sir,
>> I am working on automated test using selenium.There was initial
>> problems regarding integration with the suite with the firebug
>> extension and netexport extension (used to collect har files).I was
>> able fix the problem with minor changes in netexport code.The code
>> has been deployed from last 3 days on Delhi nodes (2G/3G).I will
>> update you on the comparisons of webpage statistics by
>> Wednesday.Also the same test will deployed at all rural locations by
>> tommorow.(already deployed at lamta)
>>
>> Things to do:
>> 1.As of now the selenium tests only download(fetching various
>> webpages),so for uploads I am working on preparing a upload test for
>> services like gmail,mediafire,dropbox,facebook and will be deployed
>> till end of the week.
>>
>> 2.Comparison of user experiences for rural location and urban
>> locations for both upload/download behaviours.
>
> --
> Aaditeshwar Seth
> Co-founder, Gram Vaani Community Media
>

> http://gramvaani.org [1]
> http://www.linkedin.com/company/gram-vaani-community-media [2]

>
> --
> You received this message because you are subscribed to the Google
> Groups "ruralnet-act4d" group.
> To unsubscribe from this group and stop receiving emails from it,

> send an email to ruralnet-act4...@googlegroups.com.
> For more options, visit https://groups.google.com/groups/opt_out [3].

>
> --
> You received this message because you are subscribed to the Google
> Groups "ruralnet-act4d" group.
> To unsubscribe from this group and stop receiving emails from it,

> send an email to ruralnet-act4...@googlegroups.com.
> For more options, visit https://groups.google.com/groups/opt_out [3].
>
>
> Links:
> ------
> [1] http://gramvaani.org
> [2] http://www.linkedin.com/company/gram-vaani-community-media
> [3] https://groups.google.com/groups/opt_out

[Aaditeshwar Seth]

The uplink tests look good, the standard deviations are low for both the
buffer sizes and the drain sizes, so looks like we have the model correct.
- What is the throughput? And since it seems steady, you should also be
able to find the buffer size in terms of time.
- You should then match this UDP throughput with what you get on TCP.
Let's see the TCP throughput is anywhere close to 3/4th of the UDP
throughput.
- And in TCP, you should also scan the trace and see if at any point
between time t and t + buffer size (in terms of time), TCP sends out
more than buffersize (in terms of bytes) data. If this happens, it would
guarantee a loss. Let's see if we can catch such cases.

The downlink does not follow a similar behavior. The buffer size seems
good since the standard deviations are low, but the pattern of emptying
out the queue at regular intervals (drain size) as on the uplink does
not seem to hold for the downlink. Can you show more downlink sequence
number traces to see if we can notice anything. You should however
anyways do the following:
- Find the throughput and see if it holds steady throughout the trace
- Compare with TCP throughput
- Do the same analysis as before, to see if in the trace you can find
any incidence of heavy download of data.

http://www.linkedin.com/company/gram-vaani-community-media

[SAMEER A PANDIT]

On 23.02.2014 12:17, Aaditeshwar Seth wrote:
> Also whatever happened to some experiments on CDN. Please go back to
> my earlier emails and send a detailed list of next steps.
>

Sir,
Regarding CDN, I am working on CDN test.On going through CDN literature
I have found that, majorly there are two type CDN services provided CDN
providers.
-Whole website hosting
In this case the origin server is hidden,so its is difficult to
find a baseline for CDN performance (performance with content hosted on
origin server).I am referring
http://dipzoom.case.edu/files/documents/measuringCDN.pdf by Sipat.This
talks about a workaround in which we request "www.foo.com/foo.jpg" with
url "www.foo.com/foo.jpg?<random_string>",the CDN edge server treats
this modified request as different url and suffers a cache miss,it
fetches the object from origin server and serves the user.This time is
more than cache hit time.The paper says if we assuming a CDN service
providers infrastructure is highly optimized,this can estimate the page
request time from origin.
I am preparing a list of CDN edge servers for various services
like akamai,limelight,edgecast,maxcdn etc.This is done using
opentraceroute servers which are located in around 22 countries(
http://lg.teleglobe.net/lg/).I am considering 6 CDN service providers
and for each provider 6-9 domains are considered which are fully severed
by CDN.
I am considering image file from each domain,and do a curl to download
the image with cahing/without caching(modified url fetch) for different
CDN edge servers collected earlier and server alloted by LDNS while
collecting tcpdump,this will give the
cahing vs no caching RTT.(sample tests show difference in cached(2ms) vs
non cached(5 ms)).

-Partial website hosting
use the same method for different objects of simillar size out of
which one is served by CDN and other by origin server.

Please let me know if test is ok, and suggest any additions if needed.

[SAMEER A PANDIT]

sir there was some mis-interpretation of timestamps in tcptrace
output,the above sample test had cached time of 351 ms and non-cached
time of 633 ms.Will update you on results from pilot tests by Friday.

[Aaditeshwar Seth]

How do you know which websites are fulled served by which CDN?

And also articulate the specific question you are trying to answer with
this experiment.

[SAMEER A PANDIT]

Sir,
Please find the update inline

On 05.03.2014 21:20, Aaditeshwar Seth wrote:
> How do you know which websites are fulled served by which CDN?

www.builtwith.com gives customers(free service gives incomplete list) of
various CDN services,I gathered this data and checked manually the ips
returned against domain names like www.kodak.com(maxcdn),
www.espncricinfo.com(akamai) etc from agni and linode servers they were
different, but with same domain names.

>
> And also articulate the specific question you are trying to answer
> with this experiment.

Below are some of specific questions to be answered.
1.How much improvement does the CDN provide in terms of RTT and
throughput?(cache vs non-cache object fetch)
2.Are CDN edge servers selected by LDNS is better than other CDN edge
servers?(IPs gathered from opentraceroute,openDNS servers)
3. As better CDN service can improve the performance significantly,which
ISPs provide better service?
4. Improvement in full page load time with CDN edge server in place for
fully served websites?

[Aaditeshwar Seth]

On 05-03-2014 22:56, SAMEER A PANDIT wrote:
> Sir,
> Please find the update inline
> On 05.03.2014 21:20, Aaditeshwar Seth wrote:
>> How do you know which websites are fulled served by which CDN?
> www.builtwith.com gives customers(free service gives incomplete list)
> of various CDN services,I gathered this data and checked manually the
> ips returned against domain names like www.kodak.com(maxcdn),
> www.espncricinfo.com(akamai) etc from agni and linode servers they
> were different, but with same domain names.

The real test will to check the ownership of these IP addresses. And
find out the actual origin IP address as well so that you can have a
baseline.

>>
>> And also articulate the specific question you are trying to answer
>> with this experiment.
> Below are some of specific questions to be answered.
> 1.How much improvement does the CDN provide in terms of RTT and
> throughput?(cache vs non-cache object fetch)
> 2.Are CDN edge servers selected by LDNS is better than other CDN edge
> servers?(IPs gathered from opentraceroute,openDNS servers)
> 3. As better CDN service can improve the performance
> significantly,which ISPs provide better service?
> 4. Improvement in full page load time with CDN edge server in place
> for fully served websites?

Didn't understand point 2, but anyways, other than quantifying the
performance in terms of latency, etc, the main agenda should be to find
out which CDN providers have partnerships with which telecom providers,
do they use DNS redirections or HTTP redirects or URL rewriting, and
whether the performance in terms of latency or page load time actually
improves with CDNs or do the CDNs only help in reducing the load on the
origin servers.

Note that the method you want to use to fetch an image twice will only
evaluate one kind of caching policy, in case it in use by the CDN. But
several other methods could be in use as well, such as the ones I listed
above -- DNS/HTTP/URL/etc. You should also document which of these
methods are in use by which combination of {CDN, telco, origin}.

And throughput will be hard to measure if the images you are downloading
are small.

[Arvind Mahla]

Sir,

Please find the updated report on Buffer Scheduling Tests attached.

Arvind 

--

You received this message because you are subscribed to the Google Groups "ruralnet-act4d" group.

To unsubscribe from this group and stop receiving emails from it, send an email to ruralnet-act4d+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

[Aaditeshwar Seth]

This looks good. The model that seems to be operating:
- Uplink: Bandwidth sharing happens because of wireless resource scheduling, common buffer leads to RTTs getting affected across simultaneous connections
- Downlink: Bandwidth sharing happens because of wireless resource scheduling, per-user buffers leads to RTTs not getting affected across simultaneous connections

The UDP Vs ICMP tests don't seem conclusive.

What is next? Still awaiting web10g analysis on TCP traces collected from rural locations to probe the connection stalls.

To unsubscribe from this group and stop receiving emails from it, send an email to ruralnet-act4...@googlegroups.com.

For more options, visit https://groups.google.com/groups/opt_out.

--
You received this message because you are subscribed to the Google Groups "ruralnet-act4d" group.

To unsubscribe from this group and stop receiving emails from it, send an email to ruralnet-act4...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Asheesh Sharma]

Sir, Please find update below:
Web10g (Graphs):
I have plotted following graphs in gnuplot for the period of Oct'13 - Feb '14:
1) Time Sequence with Current Congestion Window
2) Time Sequence with Raw round trip time(Sample RTT), RTT variance (RTTVar) and Maximum retransmit timer value (CurRto)

Preview to these graphs can be accessed in dropbox: https://www.dropbox.com/sh/lwwuu6bjwbzg0d2/8fFAC30LaR
Actual gpl files can be accessed in dropbox: https://www.dropbox.com/sh/yj9pcwvpzmmungz/rAii1I4IV8

I also analyzed few of the graphs. Please find the attachment: ReportOct13-Feb14.pdf

Pending:
1) Complete analysis of the data. Dead line: 30th march 2014

Regards,
Asheesh

[Aaditeshwar Seth]

Ashish, Please let me know on which email I can write to you.

--
You received this message because you are subscribed to the Google Groups "ruralnet-act4d" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ruralnet-act4...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Asheesh Sharma]

Sir
its asheeshkrsharma[at]me[dot]com
Kindly remove the current email address (proo...@gmail.com) from the group and please add this new one.

Regards,
Asheesh

[Rahul Nishant]

Hi Sir,
Please find attached the report containing some of the analysis on UDP
throughput etc. Currently I have done the analysis for uplink only. Can
you please suggest whether the method used to find the throughput is
correct.

Thanks & regards,
Rahul Nishant

On 24.02.2014 23:14, Aaditeshwar Seth wrote:
> The uplink tests look good, the standard deviations are low for both
> the buffer sizes and the drain sizes, so looks like we have the model
> correct.
> - What is the throughput? And since it seems steady, you should also
> be able to find the buffer size in terms of time.

Done. Did some work on oneway latency and inter arrival time along with
this. Will be doing rest of the analysis in this week.

[SAMEER A PANDIT]

Sir,
Please find the report attached for CDN.It has test for only 3 CDN
service,due to tedious work of finding appropriate objects with sizes
and there are total 10 CDN services in consideration.I am currently
working on them and will update by Friday.

[Sipat Triukose]

Hi Sameer,

May be you can look at BWget, which is a tool that can vary a TCP transfer volume from an arbitrary HTTP target, including CDN cache node.
It will ease your task of finding a suitable size object on CDN nodes. I sent you guys the code and paper about it a while ago, please let me know if you want me to send it over again.

Cheers,

Sipat

> --
> You received this message because you are subscribed to the Google Groups "ruralnet-act4d" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to ruralnet-act4...@googlegroups.com.

> For more options, visit https://groups.google.com/d/optout.
> <ReportCDN.pdf>

[Aaditeshwar Seth]

Zahir, Can you take a quick look at this. And also explain the method you used to spot connection stalls in the DEV-4 paper, ie. the specific signature of how a connection stall is different from a timeout.

-------- Original Message --------

Subject:	Re: Updates
Date:	Mon, 24 Mar 2014 15:28:04 -0700 (PDT)
From:	Asheesh Sharma <proo...@gmail.com>
Reply-To:	ruralne...@googlegroups.com
To:	ruralne...@googlegroups.com

[Aaditeshwar Seth]

I think overall this looks good, although the graphs were hard to read
with the small font size on the axes. To calculate the throughput during
the first buffer filling and the subsequent ones, you just have to look
at the rate of packet reception at the receiver. I think that's what you
have done.

Please send a comprehensive report with this UDP analysis, the earlier
TCP analysis, and other things I have suggested below, especially
contrasting the buffer size with the bandwidth x delay product.

[Aaditeshwar Seth]

This looks reasonable. Let us try to complete this asap based on the
next steps that are outlined.

Also probe for Indian websites that use CDNs, like ToI. And websites
popular in India such as Facebook.

[Rahul Nishant]

Hi Sir,
Please find attached the report on BDP, throughput comparison and
buffersize in terms of time.
I am currently combining all the reports and results of the buffer test
and will be sending the combined report in 3 days (ie by Tuesday 6th
May).

Thanks & regards,
Rahul Nishant

[SAMEER A PANDIT]

On 31.03.2014 09:20, Aaditeshwar Seth wrote:
> This looks reasonable. Let us try to complete this asap based on the
> next steps that are outlined.
>
> Also probe for Indian websites that use CDNs, like ToI. And websites
> popular in India such as Facebook.

Sir,
Please find the report attached.Regarding Indian websites most Indian
news website use Akamai CDN ,I had considered Flikart.com and
Justdial.com which are indian in Akamai,so did not considered adding TOI
or facebook.Please let me know if that will that is fine.
Regarding Throughput Comparison for CDNs I am working on bwget(tool
suggested by Sipat) to download small files multiple time over single
tcp.Please suggest if this needs to be done for all CDN services or can
be for just those which have presence in
India.(Akamai,Limelight,Amazon,Cachefly).

[Aaditeshwar Seth]

Ok, please combine and send. Ensure that the graphs are all in a
standard format with large enough font size to be readable.

--
Aaditeshwar Seth
Assistant Professor
Computer Science and Engineering
IIT Delhi

http://www.cse.iitd.ernet.in/~aseth

[Aaditeshwar Seth]

All the graphs need to be standardized.

[Rahul Nishant]

Hi Sir,
Please find attached the combined report of the buffer measurement
tests.

Thanks & regards,
Rahul Nishant

[Aaditeshwar Seth]

Folks, Can you send similar reports with standardized graphs for other
work that has been done recently:
- CDN analysis (Sameer)
- Scheduling tests (Arvind)
- Web10g and out of order tests (Arvind, Asheesh)
- Routes and roaming (Sameer/Rahul)

Anything else I am missing?

[Arvind Mahla]

Hi Sir,

Please find attached the updated report on Buffer Scheduling with standardized graphs.

Arvind

[Aaditeshwar Seth]

Arvind, What is the difference between Figs 4 and 5, and 11 and 12?

-- 
Aaditeshwar Seth
Assistant Professor
Computer Science and Engineering
IIT Delhi

http://www.cse.iitd.ernet.in/~aseth

[Arvind Mahla]

Sir,

Fig 4 & 5 are based on same dataset. Fig 4 shows phase-wise RTT on dongles (For Example: Fig 4a shows RTT  on dongle1 under different upload phases where either all dongles are idle or one or more are simultaneously uploading). Fig 5 shows the same data according to different phases on separate dongles (Fig 5a shows RTT during the Idle phase on all the three dongles).

Fig 4 is useful to make comparison of RTT across phases on a given Dongle and Fig 5 is useful to make comparison of RTT across the Dongles in a given phase.

Similar is the case with Fig 11 & 12 and they are for downlink dataset.

Arvind