PaperlayerView exact revision on arXiv

Source-linked AI summary

Collaborative Multi-bitrate Video Caching and Processing in Mobile-Edge Computing Networks

Tuyen X. Tran, Parul Pandey, Abolfazl Hajisami, Dario Pompili

arXiv:1612.01436v2cs.NI

TL;DR

The paper tackles inefficient ABR video caching in MEC networks, where multiple bitrate versions and transcoding demands must be managed under heterogeneous user and network conditions. It formulates joint collaborative caching and processing as an ILP, then proposes an online request-driven algorithm because offline optimization is impractical. Simulations report improved cache hit ratio, lower access delay, and lower external backhaul traffic than traditional approaches.

  • Problem

    ABR-aware MEC caching must select bitrate versions while accounting for transcoding relationships, storage limits, and processing capacity.

  • Method

    The paper formulates collaborative caching and processing as an ILP and proposes an online algorithm that makes placement and scheduling decisions upon each request.

  • Results

    Simulations show significant improvement over traditional approaches in cache hit ratio, content access delay, and external backhaul traffic load.

  • Takeaways & Limitations

    Collaborating MEC servers can assist one another with caching and transcoding of multi-bitrate videos.

Abstract

from arXiv · show

Recently, Mobile-Edge Computing (MEC) has arisen as an emerging paradigm that extends cloud-computing capabilities to the edge of the Radio Access Network (RAN) by deploying MEC servers right at the Base Stations (BSs). In this paper, we envision a collaborative joint caching and processing strategy for on-demand video streaming in MEC networks. Our design aims at enhancing the widely used Adaptive BitRate (ABR) streaming technology, where multiple bitrate versions of a video can be delivered so as to adapt to the heterogeneity of user capabilities and the varying of network connection bandwidth. The proposed strategy faces two main challenges: (i) not only the videos but their appropriate bitrate versions have to be effectively selected to store in the caches, and (ii) the transcoding relationships among different versions need to be taken into account to effectively utilize the processing capacity at the MEC servers. To this end, we formulate the collaborative joint caching and processing problem as an Integer Linear Program (ILP) that minimizes the backhaul network cost, subject to the cache storage and processing capacity constraints. Due to the NP-completeness of the problem and the impractical overheads of the existing offline approaches, we propose a novel online algorithm that makes cache placement and video scheduling decisions upon the arrival of each new request. Extensive simulations results demonstrate the significant performance improvement of the proposed strategy over traditional approaches in terms of cache hit ratio increase, backhaul traffic and initial access delay reduction.

I. INTRODUCTION

The paper addresses ABR video-caching challenges in MEC networks by jointly exploiting distributed storage and edge transcoding. It formulates collaborative caching and processing as an optimization problem and proposes an online request-driven strategy.

  • Motivation: MEC servers place cloud-computing, caching, and processing capabilities near mobile users at cellular base stations.Their proximity supports context-aware services and delay-sensitive applications.
  • Motivation: ABR streaming adapts video bitrate to user-device capabilities, network conditions, and requests, but storing every version independently wastes cache storage.Users with capable devices and fast connections may prefer high-resolution video, while constrained users may require lower bitrates.
  • Our vision: The proposed scheme jointly uses MEC caching and transcoding so higher-bitrate variants can be converted into lower-bitrate versions for delivery.Collaborating servers can transcode requested variants along the delivery path between the content source and home MEC server.
  • Challenges and contributions: The design must control storage overhead from multiple bitrate versions and computation demand from simultaneous real-time transcoding.These constraints motivate joint cache placement and request scheduling.
  • Challenges and contributions: The collaborative joint caching and processing problem is formulated as an ILP minimizing backhaul cost under cache-storage and processing-capacity constraints.Because the problem is NP-complete and offline approaches are impractical, the paper proposes an online algorithm that decides when each request arrives.
  • Related works: Prior multi-bitrate caching-and-processing studies consider one cache entity, while this paper targets collaboration among multiple caching and processing servers.The paper also distinguishes its online approach from methods requiring repeated optimization or accurate popularity estimates.

II. MEC CACHING SYSTEM

The MEC caching system consists of collaborating cache servers colocated with base stations, supporting storage, networking, and video transcoding. Requests can be served locally, through neighboring caches, or from the origin, with transcoding performed at different points along the path.

  • Request-service events: A request may use direct local cache delivery, local transcoding, remote retrieval, remote transcoding, or origin-server retrieval.The binary variables x, y, z, t, and w represent these mutually exclusive service events and their transcoding locations.
  • System architecture: Each MEC server is colocated with a base station and provides computation, storage, and networking within a collaborating pool.The servers are connected through backhaul links and jointly support video caching and processing.
  • System architecture: Video transcoding compresses a higher-bitrate video into a lower-bitrate version, with transcoding cost modeled as MEC-server CPU usage.Compressed-domain approaches such as bitrate and spatial-resolution reduction are identified as favorable techniques.
  • Settings: The model includes K MEC cache servers, an origin server indexed by 0, T videos, and L bitrate variants whose sizes are proportional to bitrate.A variant can be transcoded from a higher bitrate, with a specified CPU cost for the conversion.
  • Settings: Video requests arrive at each base station according to a Poisson process with rate λ_j over the evaluation period.The request set at server j is denoted N_j.
  • Settings: Each user receives data from the nearest base station, called the home BS, and each cache server has storage capacity M_j.Serving users from multiple BSs through Coordinated Multi-Point transmission is left for future investigation.

C. Backhaul Network Cost

The backhaul cost measures traffic-related resource consumption and is linked to users’ initial video playback delay. Minimizing it therefore reduces both network resource use and startup delay.

  • The cost for BS j retrieving video v_l depends on retrieval from another cache server or the origin content server.
  • Backhaul cost measures the data traffic traversing backhaul links and the corresponding network resource consumption.
  • Retrieving content through shorter paths reduces backhaul cost and directly decreases users’ initial playback delay.

III. JOINT COLLABORATIVE VIDEO CACHING AND PROCESSING

The paper formulates collaborative joint caching and processing in MEC networks as an optimization problem and presents both offline and online approaches.

  • The collaborative joint caching and processing problem is formulated, followed by an offline optimal solution and a proposed online approach.

A. Problem Formulation

The formulation jointly minimizes total backhaul serving cost by deciding cache placement and video-request scheduling under storage and processing constraints. Because the resulting ILP is NP-complete and request arrivals are unknown, the paper proposes online JCCP decisions for each arrival.

  • Problem formulation: The optimization jointly determines cache placement and video-request scheduling using available cache storage and processing resources.
  • Problem formulation: The objective is to minimize the total backhaul cost of serving all video requests.
  • Problem formulation: The constraints enforce exact variant availability, transcoding-source availability, unique request paths, cache capacity, and per-server processing capacity.
  • Problem formulation: The ILP is NP-complete, while fractional relaxation and repeated re-optimization can produce infeasible solutions, redirected requests, or wasted buffered data.
  • Problem formulation: Online JCCP uses LRU placement and makes cache-placement and request-scheduling decisions upon each new request arrival.

B. Offline Approach

The offline approach uses LRU-based cache placement and recomputes request scheduling for a long-run optimum, but its repeated integer-program complexity and request redirection make it impractical at scale.

  • Offline approach: LRU fetches uncached videos from neighboring caches or the origin, stores them at the home BS, and evicts least-recently-used entries when space is insufficient.
  • Offline approach: The offline request-scheduling problem represents videos currently being served at each BS.
  • Offline approach: The offline solution is optimal in the long run but may redirect existing requests and waste buffered data when recalculated.
  • Offline approach: Re-solving the integer program becomes highly impractical as request arrivals and the number of caching servers increase.

C. Proposed Online JCCP Algorithm

The online JCCP algorithm makes irrevocable scheduling decisions for each request, using local and neighboring caches, transcoding, processing availability, and LRU updates to serve requested videos.

  • Algorithm operation: Online JCCP makes cache placement and request scheduling decisions immediately and irrevocably when each video request arrives.It starts with empty caches and updates newly fetched videos using LRU replacement.
  • Collaborative service: If local service fails, the algorithm searches neighboring caches for the requested or a transcodable version.Neighboring-cache service can retrieve the exact version or obtain a transcodable version for processing.
  • Collaborative service: When a neighboring transcodable version exists, transcoding is assigned to the cache server or requesting server with more available processing capacity.The algorithm compares available processing resources between the two candidate servers.
  • Local service: For a request, the algorithm first serves the exact video locally, then attempts local transcoding when a suitable cached version and processing capacity exist.The local decision checks whether the requested version is cached or whether a transcodable version can be processed within the server’s capacity.
  • Fallback and updates: Requests unsatisfied by the cache system are fetched from the origin content server, while cache state is updated according to LRU policy.Origin retrieval is the fallback path and incurs the highest backhaul cost.

IV. PERFORMANCE EVALUATION

The evaluation studies joint caching and processing in a three-server MEC network with multi-bitrate videos, comparing Online-JCCP against an offline optimum and two non-equivalent baselines across three performance metrics.

  • Experimental setup: The simulation uses three MEC servers, a 1000-video library, and four bitrate variants per video.The four relative bitrates are set to 0.82, 0.67, 0.55, and 0.45 of the reference bitrate.
  • Experimental setup: Requests arrive independently at each base station according to a Poisson process, with 10,000 requests generated per base station for each simulation.Bitrate variants are selected with equal probability, and popularity distributions are shuffled across base stations.
  • Metrics: Performance is measured by cache hit ratio, average access delay, and external backhaul traffic load.Cache hit ratio includes requests satisfied by direct retrieval or transcoding.
  • Compared schemes: Online-JCCP is compared with Offline-Optimal, CachePro, and CoCache.CachePro lacks collaboration, whereas CoCache uses collaboration without transcoding and employs LRU placement.

A. Impact of cache size and processing capacities

Increasing cache size and processing capacity improves all evaluated schemes, while Online-JCCP substantially outperforms both baselines and approaches Offline-Optimal as resources become abundant.

  • Performance trends: Increasing cache size and processing capacity improves cache hit ratio, average access delay, and external backhaul traffic performance for all schemes.The comparison covers the three metrics across relative cache sizes and processing capacities.
  • Scheme comparison: Online-JCCP significantly outperforms the two baselines across a wide range of cache and processing capacities.The baselines are CachePro and CoCache from the evaluation setup.
  • Scheme comparison: At moderate cache and processing capacities, Online-JCCP performs slightly below the optimal scheme.The comparison is against Offline-Optimal.
  • Scheme comparison: At high cache sizes and processing capacities, Online-JCCP matches the optimal scheme.The performance improvement diminishes beyond a certain processing capacity, where the two schemes are almost identical.

B. Impact of request arrival rate

Online-JCCP’s cache hit ratio depends on request intensity and processing capacity, while its processing utilization also varies with arrival rate and cache capacity.

  • Cache hit ratio: Cache hit ratio decreases at high request arrival rates and low processing capacity, and increases otherwise.The relationship is reported for Online-JCCP across request arrival rates and processing capacities.
  • Processing utilization: Processing utilization increases with request arrival rate and moderate cache capacity but decreases at high cache capacity.Higher cache capacity stores more variants, leaving fewer requests that require transcoding.

V. CONCLUSIONS

The paper proposes collaborative MEC servers that jointly cache and transcode multi-bitrate videos, with performance evaluated under varying request rates, processing capacities, and cache capacities. The online JCCP scheme improves cache hit ratio, access delay, and backhaul traffic relative to traditional approaches, while approaching offline-optimal performance with abundant resources.

  • Collaborative MEC servers assist each other with caching and transcoding multi-bitrate videos.
  • Online-JCCP cache hit ratio decreases with high request arrival rates and low processing capacity, and increases otherwise.
  • Online-JCCP processing utilization increases with request arrival rate and moderate cache capacity but decreases at high cache capacity.Higher cache capacity stores more video variants, reducing requests that require transcoding.
  • The ILP minimizes total backhaul retrieval cost, while the online JCCP algorithm makes cache placement and video scheduling decisions when requests arrive.The online algorithm addresses the NP-completeness of the optimization problem and unavailable future request information.
  • JCCP improves cache hit ratio, content access delay, and external backhaul traffic over traditional approaches, with its gap from offline optimal approaching zero at high resource levels.
Loading 1612.01436v2…