[v6test] Test methodology

[Kevin Day]

(note that I'm replacing one letter of the domain name used, just to  
help those of you behind corporate content filters not get in trouble  
for this)



This is the planned visitor flow through the site - suggestions/ 
feedback greatly appreciated:


www.ipv6p*rn.com has only an A record.
v4.ipv6p*rn.com is the A only name for users without v6 support to go  
to read about the experiment and what to do to participate.
v6.ipv6p*rn.com is the AAAA only name that is for v6 users to get the  
free stuff.
v46.ipv6p*rn.com has both A and AAAA records, for a test on the front  
page.


1) A user visits www.ipv6p*rn.com. Through some javascript magic, we  
attempt to load test files from v4/v6/v46.ipv6p*rn.com.

2) Look to see which test files loaded.

   If we can load a file from v6.ipv6pr*n.com, we know v6 is working  
and invite the user to the v6 content. Skip to step 7.

   If we can load a file from v4.ipv6pr*n.com, but not from  
v46.ipv6pr*n.com - the user has v6 enabled somehow, but it's broken.  
They could load something from an address with only A records, but not  
AAAA records. Log this, let the user know about this, and direct them  
to a troubleshooting page.

   If we can load a file from v4.ipv6pr*n.com, and load a file from  
v46.ipv6pr*n.com, but not v6.ipv6pr*n.com, assume the user doesn't  
have v6 enabled.

If the user doesn't have javascript enabled, use some test images on  
the page loading from each address to let the user decide for  
themselves.

3) If they hit the v4 address published in the A record, assume they  
don't have a v6 connection and redirect them to v4.ipv6p*rn.com. Give  
them a link to an alternate host-name that only has AAAA records as a  
last resort if they DO have v6, but their resolver/browser is  
configured to use v4 over v6 even when both work. (Anyone know how  
common this is?) If they click the last-resort link, skip to step 7.

4) On their first visit to the v4 side, give the browser a unique  
tracking cookie.

5) On the v4 side, give users a 20-30MB "sample movie". Log as much  
data as possible about the transfer of this video... Average/max  
download speed, MSS/MTU, RTT, TTL, packet loss %, origin ASN, etc.  
This is large enough to get reasonably accurate statistics to rule out  
slowness due to TCP slow start, or getting unlucky with rare packet  
loss. Store this with the user's tracking cookie.

6) Present the user with instructions on how to get on ipv6. Explain  
the differences between 6to4, Teredo, tunneling, native connections,  
etc. Provide links to each, then tell the user to come back when it's  
working. User clicks a link to ipv6p*rn again, which should carry them  
on v6 from this point forward. If the user is unable to get v6  
working, they're asked to come back and click a link to notify us of  
this, with checkboxes for what methods they tried to use, and a text  
box to describe the problems they had.

7) (v6 only now) The user gets links full sized "high definition"  
resolution videos to download or watch in their browser.

8) Collect the same data about the downloads... Speeds, MSS/MTU, RTT,  
TTL, packet loss, Origin ASN, etc. Make a note of the top 48 bits of  
the source address to determine if the user is using a 6to4/Teredo  
address, a well known tunnel address, or a native v6 address. Also  
store this with the user's tracking cookie.

9) We'd also like to do an optional survey for users to answer at any  
point along the way.


 From this data recorded we can determine:

1) How many users came to look without testing v6 at all.

2) How many users tried v6 but came back to tell us that they were  
unable to get it working.

3) If given a list of v6 transitional options, which ones are users  
able to get working most often?

4) With each v6 access method, how does performance compare to v4? Are  
users penalized for download speeds if they use v6? If so, is it by  
enough that it matters?

5) What's the average MTU on v6?

6) How many users came to the site with v6 enabled, but broken? (i.e.  
what percentage of the world can't access your site if you install  
AAAA records)

7) Cross-referencing the above data with OS version, platform,  
browser, etc.



Is there any additional data that any of you would like to see us  
attempt to record? Does anyone have any technical or privacy concerns  
about what we're planning on logging now? What would you like to see  
on the survey?