The reality is that, being 52 years old and only having studied briefly in my distant past, my formal knowledge in Computing is actually lacking these days, and one subject which I know too little about, is how Push Notifications work. Back in my day, if a laptop was ‘asleep’ – i.e. In Standby – it was generally unable to be woken externally via WiFi, but did have hardware clocks that could wake it at scheduled times. Yet we know that mobile devices today, including Android and iOS devices, are able to receive push notifications from various servers, which do precisely that, and that this feature even works from behind a firewall. And so I can muse over how this might work.
I can think of two ways in which this can hypothetically work:
- The application framework can centralize the receipt of push notifications for the client device, to one UDP port number. If that port number receives a packet, the WiFi chip-set wakes up the main CPU.
- Each application that wants to receive them, can establish a client connection to a server in advance, which is to send them.
The problem with approach (1) is that, behind a firewall, by default, a device cannot be listening on a fixed port number, known to it. I.e., the same WAN IP Address could be associated with two devices, and a magic packet sent to one fixed port number, even if we know that IP Address, cannot be mapped to wake up the correct device. But this problem can be solved via UPnP, so that each device could open a listening port number for itself on the WAN, and know what its number is.
We do not always know that UPnP is available for every NAT implementation.
Approach (2) requires more from the device, in that a base-band CPU needs to keep a list, of which specific UDP ports on the client device will be allowed to wake up the main CPU, if that port receives a packet.
Presumably, this base-band CPU would also first verify, that the packet was received from the IP address, which the port in question is supposed to be connected to, on the other side, before waking the main CPU.
(Edit 12/19/2016 : Google can simply decide that after a certain Android API Number – i.e., Android version – the device needs to have specific features, that earlier Android APIs did not require.
Hence, starting from Android KitKat, or Lollipop, Google could have decided that it was no longer a special app permission, for the user to acknowledge, to wake the device. Likewise, starting from some Android version, possessing a base-band CPU might have become mandatory for the hardware, so that the API can offer a certain type of push notification.)
Also, approach (1) would have as drawback, a lack of authentication. Any networked device could just send this magic packet to any other networked device, provided that both the IP address and the port number it is sensitive to are known.
Approach (2) would bring as an advantage, that only specific apps on the client device could be enabled to receive push notifications, and the O/S would be aware of which UDP ports those are sensitive on, so that the base-band CPU would only be waking up the main CPU, if push notifications were received and associated with an app authorized to wake the device.
Also, with approach (2), the mapping of WAN port numbers back to LAN port numbers would still take place passively, through port triggering, so that the WAN-based server does not need to know, what LAN-based port number the connected port is associated with on the client device.
But, approach (2) has as a real drawback, that a server would need to keep a socket open, for every client it might want to send a push notification to. This might sound unimportant but is really not, since many, many clients could be subscribed to one service, such as Facebook. Are we to assume then, that the Facebook server also keeps one connection open to every client device? And if that connection is ever dropped, should we assume that a sea of client devices reconnect continuously, as soon as their clocks periodically wake them?