Strange Network Behavior on Chrome:Cable for SSL Handshakes
|
08-05-2014, 06:49 AM
(This post was last modified: 08-05-2014 07:08 AM by pganti.)
Post: #1
|
|||
|
|||
Strange Network Behavior on Chrome:Cable for SSL Handshakes
I am not sure if I understand totally as to how dummynet on windows work but seems like an interesting issue popped up where the initial problem was to determine whats causing SSL slowness.
First measure the same page at different speeds like for example the following: http://www.webpagetest.org/video/compare...04_3A_10J0 Specifically focus on Chrome-Cable and you see really long SSL times on the domain assets-tup.insnw.net http://www.webpagetest.org/result/140804_V4_10HX/ If I use Firefox the long SSL handshake disappears leading one to believe something is about chrome http://www.webpagetest.org/result/140804_RP_10V3/ Next I took tcpdumps on both client and server side to see whats up Serverside: https://www.cloudshark.org/captures/b9cf...0eq%201421 Clientside: https://www.cloudshark.org/captures/1f84...0eq%201421 In particular you want to focus on the following pkt numbers Serverside Pkts 37,56,59,60 Clientside Pkts 24,40,80,98 As you can see server is done within 6 ms to send all the 5 packets of interest but take around 120 ms to arrive at the client thus making it around 20-25ms spacing between the packets. What gives? The RTT between the client and server in this case is around 40 ms If there is ratelimiting on public wpt is it done based on subnet/host or some other dimension as I see that per tcp stream there seems to be effectively a rate limit of 500Kbps, is that right? |
|||
08-05-2014, 07:57 AM
Post: #2
|
|||
|
|||
RE: Strange Network Behavior on Chrome:Cable for SSL Handshakes
If we were to plot the throughput at the server side and then the same at client side you can see the attached graphs which show that server sends like 14KB in one go and then like 30KB in second go. The client however starts really small hovers around 1.*K for a while before going to 3K then 4K and ending the transfer at 5K.
Given that Chrome Cable is 5Mbps atleast 600K is permissible and particularly when the server sends out at that rate, whats stopping the client here? |
|||
08-12-2014, 03:19 AM
Post: #3
|
|||
|
|||
RE: Strange Network Behavior on Chrome:Cable for SSL Handshakes
Can you capture a net log? That should provide better detail about what the net stat sees than the traces (a lot of the net layer isn't traced right now).
|
|||
08-12-2014, 03:38 AM
Post: #4
|
|||
|
|||
RE: Strange Network Behavior on Chrome:Cable for SSL Handshakes
http://www.webpagetest.org/getgzip.php?t...netlog.txt is a sample netlog with similar long ssl behavior
|
|||
08-14-2014, 03:36 AM
Post: #5
|
|||
|
|||
RE: Strange Network Behavior on Chrome:Cable for SSL Handshakes
If you go to about://net-internals in chrome you can import the netlog and poke around. There are 2 pages you probably care about. The "timeline" page shows you bytes trabsferred over time (you'll need to zoom in quite a bit). The "events" page provides a lot more detail about what chrome is seeing, particularly around the SSL negotiation and the framing (looks like 4KB SSL frames).
It's probably best to grab netlogs and tcpdumps from the same test so they can be compared. Also, if you want to evaluate dummynet's impact you can turn it off by selecting a "native" connection. Code: t= 8852 [st= 0] +SOCKET_ALIVE [dt=8971] |
|||
08-15-2014, 08:55 AM
(This post was last modified: 08-15-2014 08:56 AM by pganti.)
Post: #6
|
|||
|
|||
RE: Strange Network Behavior on Chrome:Cable for SSL Handshakes
Code: t= 9166 [st= 314] SOCKET_BYTES_RECEIVED [@pmeenan] : You see the above problem where there is 20-30 ms spacing between the packets which were sent in the same burst from the server I would focus on this behavior using various connectivities |
|||
08-28-2014, 06:38 PM
Post: #7
|
|||
|
|||
RE: Strange Network Behavior on Chrome:Cable for SSL Handshakes
Even I observed strange behavior for Chrome and Firefox. The test which runs successfully on IE8 and IE10, doesn't terminate on Chrome and FF resulting in time-out errors.
Anyone has a solution for this |
|||
« Next Oldest | Next Newest »
|
User(s) browsing this thread: 1 Guest(s)