android push: how does it work vs battery strain?

Forums › Android › android push: how does it work vs battery strain?

Batman

Mad Scientist
29761 posts

Uber Geek

Trusted

Lifetime subscriber

#114549 23-Feb-2013 08:20

I've been trawling the net and my understanding is if you are on ios, push ing is done by a server talking to Apple then Apple remotely wakes up the iPhone to download the new stuff. My understanding is the phone is fully asleep at all times hence no battery strain.

It's that right?

It's it the same for android and does setting something to push have no effect on all the times when it is sleeping waiting for the remote poke?

Thanks

nzgeek

618 posts

Ultimate Geek

#770032 25-Feb-2013 18:15

From a low-level network standpoint, there are only a few ways that the push mechanism could be done. I can think of 3 off the top of my head:

(1) A persistent connection from the device to the server. The device makes a request for data and waits for a reply.If the server gets data it'll send it back right away, otherwise it'll eventually send back a "no data"response to keep the network connection alive.

(2) A stateless network connection using UDP. The device blips the server occasionally to say "I'm still here"and to keep any firewall holes alive.May not work with all firewalls.

(3) The device is reachable directly from the public internet. The phone doesn't need to wait for anything - it'll be contacted as needed. Doesn't work with most firewalls.

In the 2 most likely cases, the device needs to keep a network connection alive somehow. I'm not sure whether there is some sort of low power way of doing this without fully waking the CPU, but I'd say that both Android and iOS use fairly similar methods.

cyril7

9058 posts

Uber Geek

ID Verified

Trusted

Subscriber

#770039 25-Feb-2013 18:33

Hi, this is how its done, GCM replaces c2dm but the overall concept is the same.

http://developer.android.com/google/gcm/index.html

Cyril

nzgeek

618 posts

Ultimate Geek

#770258 26-Feb-2013 08:32

cyril7: Hi, this is how its done, GCM replaces c2dm but the overall concept is the same.

http://developer.android.com/google/gcm/index.html

Looks like it uses HTTP behind the scenes to transmit the data. HTTP is a request/response protocol, meaning that the server can't arbitrarily push data to the client.All our can do is delay each response as long as possible and probably respond with a 204 No Content if there's nothing to report.

As I mentioned above, this requires that the phone keeps an open network connection. I believe that the CPU can sleep while it's waiting for a response, but it still needs to wake up briefly at regular intervals to keep the connection alive. I'm not sure how long firewalls and routers keep idle TCP connection open for, but it's probably somewhere between 90 seconds and 10 minutes. Chances are that the phone has to wake up every couple of minutes, which does have an effect on battery life.

myopinion

938 posts

Ultimate Geek

#770268 26-Feb-2013 09:06

Why doesn't the phone just sleep completely and then notify the server when/if it has an IP change? That way the server always knows how to contact the phone and send email when it arrives? Hardly any battery use would be the outcome.

cyril7

9058 posts

Uber Geek

ID Verified

Trusted

Subscriber

#770269 26-Feb-2013 09:20

Hi, based on my M2M experience in the past 2min polls will keep both the mobile network and most routed networks open.

Cyril

Batman

Mad Scientist
29761 posts

Uber Geek

Trusted

Lifetime subscriber

#770285 26-Feb-2013 09:47

myopinion: Why doesn't the phone just sleep completely and then notify the server when/if it has an IP change? That way the server always knows how to contact the phone and send email when it arrives? Hardly any battery use would be the outcome.

is that how apple's push works? if so apple push superior to android push?

myopinion

938 posts

Ultimate Geek

#770291 26-Feb-2013 09:49

Not sure but that's how I would try and make it :)

Sharesies kids

Free kids accounts - trade shares and funds (NZ, US) with Sharesies (affiliate link).

nzgeek

618 posts

Ultimate Geek

#772078 1-Mar-2013 08:53

Publishing the phone's IP address isn't going to work for most WiFi networks. The vast majority of these networks (including almost every home network) use NAT.

NAT creates a private internal network where each device gets its own private IP address. These addresses are from special reserved ranges that cannot appear on the public internet. When you connect to a public network service your firewall/router forwards the data on your behalf, rewriting the address and port information to make site the data gets to the correct destination.

NAT makes it difficult to receive incoming connections. When a firewall/router receives a connection for port X, it has no way of knowing which device on the internal network to send it to. The only way to get around this is to set up port forwarding, configuring the router to send requests for specific ports to specific internal addresses. This is strictly one port to one address, so if the push server expects to contact the phone on a certain port then only one phone on the network could get push notifications.

The only way around this restriction is to get the phone to register a port number. The server can then try to connect to that port on the IP address that the registration came in on. If everything is set up correctly then out should work.

The question then is how to know what port number to register. There's no standard way to query the router for this information, and you can't expect the user to know it in every case. You could use UPnP to dynamically request a port forwarding be created, but that assumes that UPnP is supported by the router and enabled by the network administrator.

As you can see, the obvious/simple solution is anything but. It might work on cellular networks (if they aren't using NAT) but won't work on the majority of home and business networks. You also wouldn't try to use the IP address/port as a primary method and fall back to polling because then time got 2 loss of code to support. It's far better to support the one method that almost always works, even if it is less efficient.