Updates
11 views
Skip to first unread message
SAMEER A PANDIT
Dec 15, 2013, 8:47:00 AM12/15/13
to ruralne...@googlegroups.com
Sir,
Please find the updates inline.Also find the revised report attached.
On 08.12.2013 16:54, Aaditeshwar Seth wrote:
> 1. There are no references, no related work. This cannot be a research
> thesis without that
added in report.
>
> 2. I haven't seen the website in a while. But it will be good to have
> a similar graph for availability on the website.
>
> 3. Part of what you were supposed to do was to also sanitize the
> throughput (iperf/curl), latency, and other graphs. Are these scripts
> done? Would like to see the graphs being outputed, and similar graphs
> should go on to the website as well
regarding website, I am working on adding the tcptraceroute and
availability scripts to website.Will notify you ASAP,when the website is
Up and running with Availabilty status,traceroute charts.
>
> 4. I haven't read through the buffer tests since these will change.
>
> 5. Did not understand the CDN tests. What kind of CDNs are being used
> by Ebay and others? What is the criteria for ranking telcos based on
> CDNs -- are you saying that Reliance has a partnership with fewer CDNs
> than BSNL? But this is just based on your choice of 20 odd landmark
> nodes. What is the effect of using a CDN Vs not using?
Please find this in section 5.6 of report.
Effect of using CDN and not using CDN?
For getting traceroute data with not using CDN we need actual IPs of
landmarks.
For the landmarks which have CDN deployed,the DNS always gives IPs of
CDN and so exact IPs of landmarks cannot be known.One way tried was to
use a webtool to get IP for hostname,but they also give CDN IPs(checking
by whois reveals AKAMAI servers).
>
> 6. Local Vs roaming: What ISP are we talking about here? You've
> measured overall latency, hops, etc, but a direct result would be to
> compare the routes side by side -- what route is taken by packets from
> the same location but via a local SIM Vs a roaming SIM? And compare
> this across providers and a bunch of different sets of locations.
please find this in section 5.7.1,5.7.2,5.7.3
>
> 7. DNS requests: Needs more details and also figures. What is the
> quantitative impact on latency?
find this ins section 5.7.4
>
> 8. CDN performance: Give figures, can't understand what you are trying
> to say without a visual.
find this in section 5.7.5
>
>
> Please fix this and resend.
>
> Congrats on the job!
>
> On 05-12-2013 02:17, SAMEER A PANDIT wrote:
>> Sir,
>> Please find the updated report.The buffer size result is still there
>> we will send the buffer size report separately.Also the graphs are
>> made bigger in size.
>>
>> Regarding increased frequency of traceroute tests,they will be
>> performed as soon as we complete the buffer tests.
>>
>> P.S:The placement activities are over.I got placed in Oracle India.
>>
--
SAMEER A PANDIT
Mtech 2nd Year
2012MCS2821
Please find the updates inline.Also find the revised report attached.
On 08.12.2013 16:54, Aaditeshwar Seth wrote:
> 1. There are no references, no related work. This cannot be a research
> thesis without that
added in report.
>
> 2. I haven't seen the website in a while. But it will be good to have
> a similar graph for availability on the website.
>
> 3. Part of what you were supposed to do was to also sanitize the
> throughput (iperf/curl), latency, and other graphs. Are these scripts
> done? Would like to see the graphs being outputed, and similar graphs
> should go on to the website as well
regarding website, I am working on adding the tcptraceroute and
availability scripts to website.Will notify you ASAP,when the website is
Up and running with Availabilty status,traceroute charts.
>
> 4. I haven't read through the buffer tests since these will change.
>
> 5. Did not understand the CDN tests. What kind of CDNs are being used
> by Ebay and others? What is the criteria for ranking telcos based on
> CDNs -- are you saying that Reliance has a partnership with fewer CDNs
> than BSNL? But this is just based on your choice of 20 odd landmark
> nodes. What is the effect of using a CDN Vs not using?
Please find this in section 5.6 of report.
Effect of using CDN and not using CDN?
For getting traceroute data with not using CDN we need actual IPs of
landmarks.
For the landmarks which have CDN deployed,the DNS always gives IPs of
CDN and so exact IPs of landmarks cannot be known.One way tried was to
use a webtool to get IP for hostname,but they also give CDN IPs(checking
by whois reveals AKAMAI servers).
>
> 6. Local Vs roaming: What ISP are we talking about here? You've
> measured overall latency, hops, etc, but a direct result would be to
> compare the routes side by side -- what route is taken by packets from
> the same location but via a local SIM Vs a roaming SIM? And compare
> this across providers and a bunch of different sets of locations.
please find this in section 5.7.1,5.7.2,5.7.3
>
> 7. DNS requests: Needs more details and also figures. What is the
> quantitative impact on latency?
find this ins section 5.7.4
>
> 8. CDN performance: Give figures, can't understand what you are trying
> to say without a visual.
find this in section 5.7.5
>
>
> Please fix this and resend.
>
> Congrats on the job!
>
> On 05-12-2013 02:17, SAMEER A PANDIT wrote:
>> Sir,
>> Please find the updated report.The buffer size result is still there
>> we will send the buffer size report separately.Also the graphs are
>> made bigger in size.
>>
>> Regarding increased frequency of traceroute tests,they will be
>> performed as soon as we complete the buffer tests.
>>
>> P.S:The placement activities are over.I got placed in Oracle India.
>>
--
SAMEER A PANDIT
Mtech 2nd Year
2012MCS2821
Aaditeshwar Seth
Dec 20, 2013, 1:47:37 PM12/20/13
to ruralne...@googlegroups.com
Some comments.
1. In no result have you indicated how many measurements were made, over
what period, missing data that might cloud the results, etc
2. For the buffer tests, you now need to do some sanity checks. You've
calculated the buffer sizes, and you've also shown the throughput which
I'm assuming is the UDP throughput? Also show the delays and build a
table that shows the buffer size, throughput, delay, throughput x delay.
These should ideally tally.
And correlate these observations with the TCP throughput you get on
these networks for large file transfers. Get this far, and I'll then
suggest more things that should be checked to demonstrate bufferbloat.
3. The CDN tests are very confusing. You are showing the latency to the
CDN mirror of ebay.in from across different networks and locations, but
what is the baseline? You need to show something like latency to the
actual ebay.in origin server, followed by latency to its CDN mirror.
You should also explain the method you use to determine whether the IP
address you are pinging is an Akamai server or a telco server or the
origin server.
The difference between telco CDN and traditional CDN is also not clear
to me. A CDN provider like Akamai will deploy its servers inside the
telecom network, so the DNS name will be something like
ebay.in.akamai.com while a whois query for the IP address will return
the telecom provider. I think you are not clear what CDN means, please
first review the literature and then redo this analysis.
4. Section 5.7.3: What is your analysis? Or do you expect me to do this
by looking at the data dump!
5. Section 5.7.4, point 2: In the figure, are you showing results from
AP and TN SIMs used in Delhi with a Delhi SIM used in Delhi? Please
clarify these details whenever you present any data.
6. Section 5.7.5: Your results also indicate that the Delhi CDN is
faster than the TN CDN. Can you validate this by a throughput test as well.
7. Your references are not in a standard format. Some have names
capitalized, some don't even have dates for the papers.
Aaditeshwar Seth
Co-founder, Gram Vaani Community Media
http://gramvaani.org
http://www.linkedin.com/company/gram-vaani-community-media
1. In no result have you indicated how many measurements were made, over
what period, missing data that might cloud the results, etc
2. For the buffer tests, you now need to do some sanity checks. You've
calculated the buffer sizes, and you've also shown the throughput which
I'm assuming is the UDP throughput? Also show the delays and build a
table that shows the buffer size, throughput, delay, throughput x delay.
These should ideally tally.
And correlate these observations with the TCP throughput you get on
these networks for large file transfers. Get this far, and I'll then
suggest more things that should be checked to demonstrate bufferbloat.
3. The CDN tests are very confusing. You are showing the latency to the
CDN mirror of ebay.in from across different networks and locations, but
what is the baseline? You need to show something like latency to the
actual ebay.in origin server, followed by latency to its CDN mirror.
You should also explain the method you use to determine whether the IP
address you are pinging is an Akamai server or a telco server or the
origin server.
The difference between telco CDN and traditional CDN is also not clear
to me. A CDN provider like Akamai will deploy its servers inside the
telecom network, so the DNS name will be something like
ebay.in.akamai.com while a whois query for the IP address will return
the telecom provider. I think you are not clear what CDN means, please
first review the literature and then redo this analysis.
4. Section 5.7.3: What is your analysis? Or do you expect me to do this
by looking at the data dump!
5. Section 5.7.4, point 2: In the figure, are you showing results from
AP and TN SIMs used in Delhi with a Delhi SIM used in Delhi? Please
clarify these details whenever you present any data.
6. Section 5.7.5: Your results also indicate that the Delhi CDN is
faster than the TN CDN. Can you validate this by a throughput test as well.
7. Your references are not in a standard format. Some have names
capitalized, some don't even have dates for the papers.
Co-founder, Gram Vaani Community Media
http://gramvaani.org
http://www.linkedin.com/company/gram-vaani-community-media
Rahul Nishant
Jan 6, 2014, 3:40:54 PM1/6/14
to ruralne...@googlegroups.com
Hi Sir,
Please find the updates below :
- The coding for the update mails have been completed. Currently I have
set it for testing using a personal email id. The update to the ruralnet
group will start in 3 days ( on 9th Jan,2014).
- Task : Will be correlating the TCP thoughputs, delay and buffer sizes.
The TCP download tests for new nodes need to be done for this purpose.
> 2. For the buffer tests, you now need to do some sanity checks. You've
> calculated the buffer sizes, and you've also shown the throughput which
> I'm assuming is the UDP throughput? Also show the delays and build a
> table that shows the buffer size, throughput, delay, throughput x delay.
> These should ideally tally.
> And correlate these observations with the TCP throughput you get on
> these networks for large file transfers. Get this far, and I'll then
> suggest more things that should be checked to demonstrate bufferbloat.
The throughput graph shown in the report was the TCP download throughput
taken from the thesis of Goutham and Gaurav, not the UDP throughput.
Thanks & regards,
Rahul Nishant
> Some comments.
>
> 1. In no result have you indicated how many measurements were made, over
> what period, missing data that might cloud the results, etc
>
>
> --
> 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.
>
Please find the updates below :
- The coding for the update mails have been completed. Currently I have
set it for testing using a personal email id. The update to the ruralnet
group will start in 3 days ( on 9th Jan,2014).
- Task : Will be correlating the TCP thoughputs, delay and buffer sizes.
The TCP download tests for new nodes need to be done for this purpose.
> 2. For the buffer tests, you now need to do some sanity checks. You've
> calculated the buffer sizes, and you've also shown the throughput which
> I'm assuming is the UDP throughput? Also show the delays and build a
> table that shows the buffer size, throughput, delay, throughput x delay.
> These should ideally tally.
> And correlate these observations with the TCP throughput you get on
> these networks for large file transfers. Get this far, and I'll then
> suggest more things that should be checked to demonstrate bufferbloat.
taken from the thesis of Goutham and Gaurav, not the UDP throughput.
Thanks & regards,
Rahul Nishant
> Some comments.
>
> 1. In no result have you indicated how many measurements were made, over
> what period, missing data that might cloud the results, etc
>
>
> 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.
>
Aaditeshwar Seth
Jan 6, 2014, 10:21:47 PM1/6/14
to ruralne...@googlegroups.com
On point 2, showing TCP throughput is meaningless. There are too many
convoluting processes within TCP. You need to compare with UDP
throughput, and only as a second level argument or more as a symptom you
should show TCP throughput.
convoluting processes within TCP. You need to compare with UDP
throughput, and only as a second level argument or more as a symptom you
should show TCP throughput.
rahul nishant
Jan 7, 2014, 7:17:51 AM1/7/14
to ruralne...@googlegroups.com
Hi Sir,
We have currently done throughput tests using TCP only. We will need to modify the download and upload tests to use UDP instead. So, I will be working on that now. Curl does not allow UDP downloads, so I will try with Iperf and also try to go through some other tools.
For more options, visit https://groups.google.com/groups/opt_out.
--
Aaditeshwar Seth
Co-founder, Gram Vaani Community Media
http://gramvaani.org
http://www.linkedin.com/company/gram-vaani-community-media
--
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.
SAMEER A PANDIT
Jan 16, 2014, 9:14:45 PM1/16/14
to ruralne...@googlegroups.com
Sir,
Please find the updates inline and attached Roaming vs Local report
IP of origin server could not be found.Thus there is no baseline so I
think CDN analysis will require a separate test rather than using
tcptraceroute test data.I will update regarding the test.
- Throughput analysis for Roaming and Local.
- User Experience at rural locations.
Regarding this I have found a browser automation tool (Selenium)
,which can used emulate browsers , thus a user pattern of browsing can
emulated and comparison in User exp. perceived at deployed location for
various ISP (rural vs urban) can be done.
Please find the updates inline and attached Roaming vs Local report
> 3. The CDN tests are very confusing. You are showing the latency to
> the CDN mirror of ebay.in from across different networks and
> locations, but what is the baseline? You need to show something like
> latency to the actual ebay.in origin server, followed by latency to
> its CDN mirror.
>
> You should also explain the method you use to determine whether the IP
> address you are pinging is an Akamai server or a telco server or the
> origin server.
In the CDN test performed,the DNS always resolved the IP of a CDN.So the
> the CDN mirror of ebay.in from across different networks and
> locations, but what is the baseline? You need to show something like
> latency to the actual ebay.in origin server, followed by latency to
> its CDN mirror.
>
> You should also explain the method you use to determine whether the IP
> address you are pinging is an Akamai server or a telco server or the
> origin server.
IP of origin server could not be found.Thus there is no baseline so I
think CDN analysis will require a separate test rather than using
tcptraceroute test data.I will update regarding the test.
>
> The difference between telco CDN and traditional CDN is also not clear
> to me. A CDN provider like Akamai will deploy its servers inside the
> telecom network, so the DNS name will be something like
> ebay.in.akamai.com while a whois query for the IP address will return
> the telecom provider. I think you are not clear what CDN means, please
> first review the literature and then redo this analysis.
Things to do:
> The difference between telco CDN and traditional CDN is also not clear
> to me. A CDN provider like Akamai will deploy its servers inside the
> telecom network, so the DNS name will be something like
> ebay.in.akamai.com while a whois query for the IP address will return
> the telecom provider. I think you are not clear what CDN means, please
> first review the literature and then redo this analysis.
- Throughput analysis for Roaming and Local.
- User Experience at rural locations.
Regarding this I have found a browser automation tool (Selenium)
,which can used emulate browsers , thus a user pattern of browsing can
emulated and comparison in User exp. perceived at deployed location for
various ISP (rural vs urban) can be done.
Aaditeshwar Seth
Jan 18, 2014, 1:42:49 PM1/18/14
to ruralne...@googlegroups.com
CDN: You can find the origin server's IP by resolving from some other
DNS server. You can either set your DNS server to a public server in the
US for example, or you can use one of the open traceroute servers to
resolve the domain.
DNS server. You can either set your DNS server to a public server in the
US for example, or you can use one of the open traceroute servers to
resolve the domain.
Reply all
Reply to author
Forward
0 new messages