[v6test] Test download server

Tue Mar 25 01:49:09 UTC 2008

Kevin Day wrote:
> I'm starting to do some performance/load testing of our download  
> cluster. If anyone wants to see what kind of speeds they're getting  
> from us right now, there are a couple of videos posted at http://testdownload.ipv6experiment.com 
> . Accessible both over v4 and v6, to help with speed comparisons.
>
> If you're getting really bad or really good download speeds off that,  
> let me know off list. I'm at the point where I can start looking for  
> routing/performance problems to get them fixed before it launches.
>
> Teredo users may get some really bad performance right now, this is a  
> known problem and I'm working on a solution.
>
> Right now I'm just testing our monitoring system. Each download is  
> logging things like MSS, packet loss, window size, etc. I'm not too  
> worried about problems with the video compression or content itself  
> yet, that comes next!
>
> Don't spread the URL all over the place, please. :)
>
> -- Kevin
>
> _______________________________________________
> V6test site list
> V6test at lists.your.org
> http://mail.your.org/mailman/listinfo/v6test
>   
This test was run from a box off of a HE.net 6in4 tunnel in San 
Francisco, California - grafenberg.thejof.com ( 69.59.173.73 / 
2001:470:1f05:1f6::1 )

First, the v6 test:

jonathan at grafenberg:~$ traceroute -6 testdownload.ipv6experiment.com
traceroute to testdownload.ipv6experiment.com 
(2001:4978:8:1:218:8bff:fe52:b5b0), 30 hops max, 40 byte p         ackets
 1  xxx.tunnel.tserv3.fmt2.ipv6.he.net (2001:470:1f04:xxx::x)  5.956 ms  
7.629 ms  6.481 ms
 2  1g-3-20.core1.fmt2.ipv6.he.net (2001:470:0:45::1)  13.699 ms  13.685 
ms  13.680 ms
 3  10g-1-1.core1.sjc2.ipv6.he.net (2001:470:0:31::2)  8.260 ms  8.235 
ms  8.215 ms
 4  10g-1-1.core1.chi1.ipv6.he.net (2001:470:0:3c::2)  69.525 ms  69.508 
ms  69.608 ms
 5  2001:470:0:7f::2 (2001:470:0:7f::2)  65.961 ms  65.941 ms  65.923 ms
 6  unassigned.v6.your.org (2001:4978:8:1:218:8bff:fe52:b5b0)  65.269 
ms  61.635 ms  61.476 ms

jonathan at grafenberg:~$ time wget -6 -O /dev/null 
http://testdownload.ipv6experiment.com/24-h264-aac.mp4
--xx:xx:24--  http://testdownload.ipv6experiment.com/24-h264-aac.mp4
           => `/dev/null'
Resolving testdownload.ipv6experiment.com... 
2001:4978:8:1:218:8bff:fe52:b5b0
Connecting to 
testdownload.ipv6experiment.com|2001:4978:8:1:218:8bff:fe52:b5b0|:80... 
connected.
HTTP request sent, awaiting response... 200 OK
Length: 36,228,544 (35M) [application/octet-stream]

100%[==...==>] 36,228,544   800.98K/s    ETA 00:00

xx:xx:56 (1.07 MB/s) - `/dev/null' saved [36228544/36228544]

real    0m32.312s
user    0m0.076s
sys     0m0.216s

And the v4 test:

jonathan at grafenberg:~$ traceroute testdownload.ipv6experiment.com
traceroute to testdownload.ipv6experiment.com (216.14.100.2), 30 hops 
max, 40 byte packets
 1  xxx (xxx.xxx.xxx.xxx)  18.298 ms  18.359 ms  18.340 ms
 2  xxx (xxx.xxx.xxx.xxx)  0.562 ms  0.792 ms  0.775 ms
 3  ge-6-xx.car2.SanFrancisco1.Level3.net (4.xxx.xxx.xxx)  0.735 ms  
0.721 ms  0.705 ms
 4  ae-4-4.ebr2.SanJose1.Level3.net (4.69.133.158)  4.533 ms  4.523 ms  
4.489 ms
 5  ae-3.ebr1.Denver1.Level3.net (4.69.132.58)  27.187 ms  27.180 ms  
27.165 ms
 6  ae-1-100.ebr2.Denver1.Level3.net (4.69.132.38)  36.869 ms  36.562 
ms  36.295 ms
 7  ae-3.ebr3.Chicago1.Level3.net (4.69.132.62)  58.117 ms  58.111 ms  
58.085 ms
 8  ae-68.ebr1.Chicago1.Level3.net (4.69.134.57)  51.561 ms  63.141 ms  
63.129 ms
 9  ae-23-52.car3.Chicago1.Level3.net (4.68.101.39)  52.352 ms 
ae-13-53.car3.Chicago1.Level3.net (4.68.101.71)  78.828 ms 
ae-23-54.car3.Chicago1.Level3.net (4.68.101.103)  52.284 ms
10  261.xe-0-3-0.cr2.ord1.us.scnet.net (4.71.100.26)  57.616 ms  57.756 
ms  57.703 ms
11  v21.ar1.ord1.us.scnet.net (216.246.95.243)  57.688 ms 
v21.ar2.ord1.us.scnet.net (216.246.95.244)  57.615 ms 
v21.ar1.ord1.us.scnet.net (216.246.95.243)  58.066 ms
12  sce.bg0.ord01.gw.stilenet.com (64.202.111.33)  57.536 ms 
sce.bg0-v4.ord01.gw.stilenet.com (64.202.111.37)  57.524 ms 
sce.bg0.ord01.gw.stilenet.com (64.202.111.33)  57.544 ms
13  testdownload.ipv6experiment.com (216.14.100.2)  57.737 ms  57.691 
ms  57.652 ms

jonathan at grafenberg:~$ time wget -O /dev/null 
http://testdownload.ipv6experiment.com/24-h264-aac.mp4
--xx:xx:15--  http://testdownload.ipv6experiment.com/24-h264-aac.mp4
           => `/dev/null'
Resolving testdownload.ipv6experiment.com... 216.14.100.2, 
2001:4978:8:1:218:8bff:fe52:b5b0
Connecting to testdownload.ipv6experiment.com|216.14.100.2|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 36,228,544 (35M) [application/octet-stream]

100%[==...==>] 36,228,544     3.90M/s    ETA 00:00

17:43:25 (3.85 MB/s) - `/dev/null' saved [36228544/36228544]

real    0m9.087s
user    0m0.080s
sys     0m0.196s

Despite more hops over v4, over time, the latency still seems ever so 
slightly lower via v4.
v4 averages around 55-59 ms, and v6 more like 60-65ms.

Traffic through HE.net seems to be consistently rate-limited 1 
Megabit/s. v4 transit caps out around 5 Megabits/s.

Application layer performance seems ok. Over v6, I noticed a couple of 
dropped ACKs that caused some the occasional fast restart. Here's a 
graph for raw layer 4 (rather than TCP payload) throughput per unit 
time: http://thejof.com/pub/caps__your.org_v6throughput.jpg

Does anyone else find it funny that others only used the "24" video in 
their examples?

Cheers,
jonathan