Dormancy

Zorog · EVDO Heavy User Joined: 08 Jul 2005 Posts: 135

I noticed while reading these forums that some folks think dormancy is a problem. It doesn't have to be.

Dormancy is the process where your AT (access terminal) will release the RF resources after 20 seconds of inactivity. (Higher quality devices will release the RF sooner.) Most important for you to know - your session is still alive on the PDSN and the switching infrastructure. You are simply re-connected to the RF resources you need when you need them. As an illustration - a cell site may have hundreds of phones registered in a sector, but only a limited number of them can place a call simultaneously. They are all there communicating to the site, but RF resources are only used when a call actively needs to be made.

It is VITAL to the network that devices go dormant when not actively transmitting data. Just like your cell phone, you want resources available to you every time you want to make a call. Imagine how it would be if everyone refused to hang up their call - a small number of users would tie up the system, and you would be out of luck. (Remember the "fast busy" signal we used to get with analog? - OK you young pups, that wasn't so long ago!)

Waking up from dormancy in EVDO only takes between 1.5 and 2 seconds, it is quite fast. Waking up from dormancy in 1X takes about 3 to 4.5 seconds. THIS is the problem.

The TCP/IP stack used by Microsoft has the default TCPInitialRTT timer set to 3 seconds, so your first packet may time-out before you can finish waking up from dormancy. You can see this when you are browsing - go dormant, click on a link, and get "page not found", hit refresh, and it works. This is because your first packet died. By the way, if you cannot reproduce this - you have no problem with dormancy.

To correct the Microsoft stack, and allow dormancy to work smoothly, check out the following registry keys:

http://support.microsoft.com/?kbid=170359

SPECIAL NOTE - Do not modify the TCPMaxDataRetransmissions - it will slow down your overall experience in a wireless environment. The first TCP packet is only sent once, so upping the retransmissions is irrelevant for dormancy issues.

Conclusion - DO NOT keep a session alive with "keep-alive" methods - you are only going to hurt yourself and the technology in the long run!

Michael · Posted: Fri Jul 15, 2005 1:28 pm Post subject:

Zorog,

Great post. I am going to link to this post over at EVDOinfo.com, on the pages that talk about Dormancy and Dormant killing.
_________________
EVDO :: EVDO News :: EVDO Antennas :: Buy Verizon :: Buy Sprint :: EVDO Amplifier

brendanhoar · EVDO Heavy User Joined: 09 Jun 2005 Posts: 140

Zorog - until I configured my ssh client to send keepalive packets, I was routinely losing the ssh connection due to dormancy-related timeouts.

In your (experienced) opinion, is it better for me (and/or other users) to set-and-forget the ssh client (putty) to use keepalive packets, or change Window's networking parameters?

-brendan

Zorog · EVDO Heavy User Joined: 08 Jul 2005 Posts: 135

The windows TCP fix should correct it. Looking at some PuTTY support pages, the product does not seem to have it's own timeout values, so the issues is most likely at the TCP layer.

frissell · EVDO Fledgling Joined: 23 Jun 2005 Posts: 22

But it happens that my (updated) installation of Windows 2000 Professional Edition doesn't have an entry called TCPInitialRTT so I can't reset it.

Also, you didn't recommend what to set it to.

Zorog · EVDO Heavy User Joined: 08 Jul 2005 Posts: 135

Ouch! You're right - I neglected to recommend a new value! The key must be added (so you won't find it to edit). The instructions are in the link above.

I recommend setting it to 5 seconds instead of the 3 seconds currently the default. For some versions of Windows, set it to "5", others set to 5000, as it's in milliseconds. The instructions are pretty clear.