Computer Engineering is coming soon

We're still writing and reviewing the notes for every subject, and we'll open it when every subject is ready. Meanwhile you can practise aptitude, ask Aria about any subject, or see the branches that are ready.

Mobile, multimedia and energy-aware clouds

Mobile, multimedia and energy-aware clouds: concepts, models and calculations for offloading, multimedia QoS and energy-efficient cloud designs for mobile devices and streaming applications.

Drafted with Aria, reviewed by the AiCanCode.org team · Updated . How we write and check our notes. Spotted an error? Use Give Feedback at the bottom of the page.

Why it matters

Mobile devices have limited battery and CPU resources. Engineers use mobile and energy-aware cloud techniques to decide when to run tasks on the device, when to offload to edge or cloud servers, and how to stream multimedia with required quality while saving energy and meeting latency and bandwidth constraints.

Key ideas

Start from first principles: work, energy, power and information transfer.

  • Work/Energy: energy is the capacity to do work; in computing we account for electrical energy consumed by CPU, radio and display, measured in joules (J).
  • Power: rate of energy use, measured in watts (W). Relationship: energy = power × time.
  • Bandwidth: the data rate of a communication link, measured in bits per second (bps). Higher bandwidth reduces transmission time for fixed data size.
  • Latency: the time delay between sending a request and receiving a response, measured in seconds (s). For offloading, total latency includes transmission, propagation, queuing and remote processing delays.
  • Offloading decision: moving computation from a mobile device to a remote server. Formally, an offloading decision is beneficial if energy saved on the device and/or performance improvement outweighs communication costs (energy + latency).

Define formally:

  • Local execution energy: energy consumed by running an application on the mobile CPU. Denote E_local.
  • Offloading energy: energy consumed to transmit inputs + receive outputs + any residual local processing. Denote E_offload.
  • Energy-aware cloud: cloud systems and schedulers designed to reduce total energy consumption (device + cloud datacentre) or prioritize device battery life.

Assumptions commonly used in exam derivations:

  • Computation cost scales linearly with number of CPU cycles required; denote cycles C and energy per cycle e_c.
  • Transmission energy scales with data size D (bits) and energy per bit e_b; transmission time = D / B where B is bandwidth. In practice include radio tail energy/state transitions (3G/4G/5G and Wi‑Fi) that add overhead after a transfer.
  • Remote computation time on cloud is often much smaller than mobile, but add network RTT (round trip time). When edge servers are used, assume smaller latency than central cloud.

Derivation commonly asked: condition for offloading to save device energy. Step 1: write energies.

  • Local: E_local = C · e_c.
  • Offload: E_offload = D·e_b_tx + D_out·e_b_rx + E_local_res, where D is input bits, D_out is output bits returned, e_b_tx and e_b_rx are energy per bit for transmit and receive, and E_local_res is any residual local work. Step 2: offloading beneficial if E_offload < E_local. Step 3: substitute and rearrange: C·e_c − E_local_res > D·e_b_tx + D_out·e_b_rx.

Diagram to draw for exam answers (describe exactly):

  • Draw three boxes left-to-right labelled "Mobile device", "Wireless link (cell/Wi-Fi)" and "Cloud/Edge server". Draw arrow from Mobile to Cloud labelled "D bits, bandwidth B (bps), latency t_net (s)" and return arrow for outputs labelled "D_out bits". Near Mobile box draw CPU icon and label power P_cpu (W) and battery symbol. Near Wireless link label energy per bit e_b (J/bit).

Multimedia-specific concepts:

  • Bitrate: bits per second required by a video stream.
  • Adaptive Bitrate (ABR) streaming: server/client adjust bitrate to current network conditions to trade quality vs. stall.
  • Jitter: variation in packet arrival time; causes playback glitches. Play-out buffer is used to absorb jitter, but increases end-to-end latency.
  • Requirements: interactive video (video call) needs low one-way latency (<150 ms; round-trip <300 ms per ITU-T G.114), while streaming on-demand tolerates higher latency but requires steady bandwidth.

Energy-aware cloud techniques used in practice:

  • Computation offloading (full or partial/method-level).
  • Workload consolidation in cloud datacenters (to turn off idle servers), VM migration to fewer hosts.
  • Use of edge servers (fog/edge computing) to reduce network latency and, when responses are smaller or tail time is reduced, the device’s radio on-time (but transmit energy per bit is set by the wireless link).
  • Multimedia transcoding at edge to reduce downstream bandwidth and device decoding energy.

Where to trade off: there is an energy-delay trade-off: minimizing energy often increases latency (e.g., by batching transmissions), while minimizing latency increases energy (e.g., keeping radios on high-power states).

Formulas

E = P · t — E is energy in joules (J), P is power in watts (W), t is time in seconds (s). Applies to any steady power usage over time. Battery energy from capacity: E_batt[J] ≈ V_nominal[V] · Capacity_Ah · 3600. E_local = C · e_c — C is CPU cycles (cycles), e_c is energy per cycle (J/cycle). Use when computation energy scales with cycles. At chip level, P_dynamic ≈ C_eff · V^2 · f, so DVFS changes both e_c (∝ V^2) and t_comp (∝ 1/f). t_comp = C / f — t_comp is computation time (s), C is cycles, f is CPU frequency (cycles/s or Hz). Use for single-core deterministic CPU. E_tx = D / B · P_tx = t_tx · P_tx — E_tx is energy to transmit data D (bits) at bandwidth B (bps) using radio power P_tx (W); t_tx is transmit time (s). E_offload = E_tx + E_rx + E_local_res — sum of transmit, receive and residual local energy; units J. Throughput requirement b = frame_rate · bits_per_frame — b in bps; apply for video stream bitrate calculation. Latency_total = t_uplink + t_server + t_downlink + t_processing — total delay experienced in offloading or interactive multimedia; where t_uplink = t_tx_up + t_prop_up + t_queue_up, t_server is remote compute time, t_downlink has analogous components, and t_processing is local encode/decode or residual processing. Energy_saving_if_offload = E_local - E_offload — positive means offloading saves device energy.

(For each symbol: J = joule, W = watt, s = second, Hz = cycles/s, bps = bits per second.)

Worked examples

Example 1 — Energy offloading numeric (step-by-step) Problem: A mobile app requires C = 4·10^9 CPU cycles. Mobile CPU frequency f = 1 GHz. Mobile CPU active power P_cpu = 1.2 W. Assume energy per cycle e_c = P_cpu / f = 1.2 / 1e9 = 1.2e-9 J/cycle. Data to send D = 5·10^6 bits, output D_out = 1·10^6 bits. Radio transmit power P_tx = 0.8 W, receive power P_rx = 0.5 W, bandwidth B = 2·10^6 bps. Step 1: E_local = C · e_c = 4e9 · 1.2e-9 = 4.8 J. (Formula used: E_local = C·e_c) Step 2: t_tx = D / B = 5e6 / 2e6 = 2.5 s. t_rx = D_out / B = 1e6 / 2e6 = 0.5 s. Step 3: E_tx = t_tx · P_tx = 2.5 · 0.8 = 2.0 J. E_rx = 0.5 · 0.5 = 0.25 J. Step 4: Assume negligible local residual work E_local_res = 0. Total E_offload = 2.0 + 0.25 = 2.25 J. Step 5: Energy saving = E_local − E_offload = 4.8 − 2.25 = 2.55 J. Conclusion: offloading saves 2.55 J on the device.

Example 2 — Multimedia bitrate and buffering (numeric) Problem: A streamed video is 30 fps, each frame compressed to 100 kB. Calculate bitrate and the buffer size (in seconds) needed to hold 5 MB of data. Step 1: bits per frame = 100 kB × 8 = 800 kbits = 8e5 bits. Step 2: bitrate b = 30 × 8e5 = 24e6 bps = 24 Mbps. (Formula: b = frame_rate · bits_per_frame) Step 3: buffer size seconds = buffer_bytes · 8 / b = (5e6 · 8) / 24e6 = 40e6 / 24e6 = 1.666... s. Buffer holds ~1.67 seconds of video.

Example 3 — Model answer (theory): Explain energy-aware offloading policy. Answer must state: define local vs offload energies (formulas), show decision condition E_offload < E_local, discuss latency constraint (Latency_total ≤ Latency_budget), describe trade-offs and role of edge servers to reduce t_server and E_tx, and mention practical factors: variability of wireless channel, user mobility, security, and cloud-side energy cost if total system energy is considered.

Common mistakes

  • Forgetting to include both transmit and receive energy when offloading.
  • Using bytes vs bits inconsistently (mixing B and b). Convert bytes to bits (×8).
  • Omitting network latency in offloading time and concluding incorrectly on feasibility.
  • Using CPU cycles but forgetting to convert cycles/s (frequency) to time when needed.

For your exam

Possible question types:

  • Definitions: give clear definitions of offloading, bandwidth, latency, energy-aware cloud.
  • Short notes: describe ABR streaming, edge computing benefits for mobile, or energy-delay trade-off.
  • Derivations: derive the offloading decision inequality from energy models (show algebra). Include assumptions.
  • Numericals: compute energy saved by offloading with given P and times, compute buffer duration from data size and bitrate.
  • Diagrams: draw mobile–wireless–edge/cloud architecture and label flows, power and latency. Full-marks answers include: correct formulas with units, clear step-by-step calculations, stated assumptions (e.g., constant bandwidth), and a concluding sentence interpreting results.

Quick check

  1. What is the condition (inequality) for offloading to save device energy?
  2. How do you compute transmit time t_tx from data size D and bandwidth B?
  3. If a video stream has 24 Mbps bitrate, how many seconds does a 12 MB buffer hold? Answers: 1) E_offload < E_local. 2) t_tx = D / B (D in bits, B in bps). 3) 12 MB = 12·10^6 bytes = 96·10^6 bits; time = 96e6 / 24e6 = 4 s.

Finished this topic? Mark it so your progress, study plan and readiness keep up.

Stuck on something here?