Re:Let's not use HTTP over TCP for mobile apps anymore
by chetanahuja on Friday April 10, 2015 @01:42PM
(#9355895)
Just to get the pricing thing out of the way, the pricing is not to license a protocol but to actually distribute app content -- from our servers worldwide. And our current customers are finding that it's pretty comparable to what they maybe already paying for such a service to other providers.
Agreed that the standards will get there at some point. And it'll be a joyous day for all. But I'll have to humbly disagree with you that TCP Fast Open or Multipath or HTTP/2 provide the answer for the mobile app use case today.
TFO will save you one round-trip for the handshake.. but that's about it.
Multipath is only available for Apple's own use the last I heard so it's not really a factor... but even if it does become available, it's still TCP.
And in both cases, TCP is still hampered by the usual problems outlined in the blogpost. No concept of a session, slow-start, exponential backoff, unrealistically long timeouts etc. etc.
I guess the meta point here is that TCP is designed for the general case and it's great for what it is meant for. But by narrowing the use-case to native mobile apps, one can constrain the problem domain and gain more freedom in what you can design.
Agreed that the standards will get there at some point. And it'll be a joyous day for all. But I'll have to humbly disagree with you that TCP Fast Open or Multipath or HTTP/2 provide the answer for the mobile app use case today.
TFO will save you one round-trip for the handshake.. but that's about it.
Multipath is only available for Apple's own use the last I heard so it's not really a factor... but even if it does become available, it's still TCP.
And in both cases, TCP is still hampered by the usual problems outlined in the blogpost. No concept of a session, slow-start, exponential backoff, unrealistically long timeouts etc. etc.
I guess the meta point here is that TCP is designed for the general case and it's great for what it is meant for. But by narrowing the use-case to native mobile apps, one can constrain the problem domain and gain more freedom in what you can design.