Response time, as defined in this paper, refers to the round-trip time (RTT) for mobile clients connected to cloud services and carrier networks. The authors consider the challenge of supporting two specific mobile use cases: (1) mobile gaming, which is highly interactive, and (2) mobile desktop applications such as word processing and PowerPoint slide shows. They conclude that current cloud mobile architectures cannot provide satisfactory performance, and discuss some possible techniques for addressing response time.

The proposals make certain assumptions. For example, one technique assumes that the applications rely on cloud Internet-as-a-service (IaaS) and platform-as-a-service (PaaS) for computation, meaning that the mobile device is lightweight and is used to send brief commands upstream and to receive changing video or screen images in the downstream direction.

For related techniques, the authors consider frame rate adaptation (used in video streaming) and interaction with a Citrix server (for remote desktop viewing), and conclude that performance for these is unacceptable. Therefore, other techniques must be considered:

• Uplink delay optimization: the authors have developed an enhanced protocol based on the standard user datagram protocol (UDP) to control retransmission at the application layer. This is specific to an application the authors created and is probably not a totally original idea.
• Downlink delay reduction: this addresses the need for adaptive reduction of bandwidth based on the distance and texture of least important visual objects (joint rendering and encoding adaptation (JREA)). Impressive-looking round-trip improvements are indicated in figure 10, but the authors skip very lightly over a host of very difficult issues. For example, in one of the “improvements” shown in figure 9b, all of the buildings in the scene have disappeared, and in figure 9d, the buildings are displayed with extremely coarse resolution. How does the application prioritize what gets viewed and which textures are of highest priority? How does JREA apply both criteria at the same time (that is, will the buildings be absent or blurry, or some combination of these)? Since these adaptations are made to keep the end user happily playing the game, do these changes impact the player’s ability to play the game? As written, this reads like a technical trick.
• Computing and storage at the edge of the mobile network: this calls for adding new services to carrier networks to replay Internet cloud services. It is true that moving game or view servers out to the edge of the carrier network would reduce RTT, but there is no discussion of how a multiplayer game could be distributed across a wireless network. For example, figure 13 shows a single mobile user connected to “Mobile Cloud 1.” How does this work if other players are connected to mobile clouds at other gateways? At one point on page 9, the authors suggest that the mobile cloud network could be extended to include the mobile devices themselves, suggesting that the mobile devices are not acting as thin clients anymore. Doesn’t this contradict the original constraints?
• Mobile cloud scheduling: the authors suggest that the cloud could be scheduled akin to resources on a high-performance cluster (HPC). This is a very difficult problem.

The main contributions in this paper include measurements of response time under congested network conditions for two specific types of interactive wireless applications, and a breakdown of the elements or components contributing to the observed RTT. New approaches are only suggested or are very preliminary. The authors do a reasonable job of explaining some of the major challenges.

It is interesting to consider other real-time or near real-time mobile applications, such as traffic maps and location updaters. These apparently operate to the user’s satisfaction today, so how do the characteristics of these applications differ from the ones described in the paper? There is also literature regarding real-time requirements for sensor networks [1], which might be explored.

The paper is generally well written, but could certainly have used some editing by a native English speaker. Many sentences are grammatically correct but awkward, such as, “To address this, we can take the help of encoding adaptation.” Basic proofing for typographical errors and incorrect punctuation would also have helped.