Reliability is one of the most prominent issues in edge computing due to wide geographical distribution, limited resource capacities, and dominant wireless connectivity. As a consequence, edge computing is affected by limited hardware redundancy, a lack of sophisticated support systems (e.g., power backup, efficient cooling), and unstable and intermittent wireless connections, making reliability difficult to achieve. It is harder to attain the same reliability levels as in cloud data centers, which do not suffer from these problems due to centralized architecture and resource abundance.

Reliability in Edge Offloading

Reliability problems in edge computing extend to edge offloading as well. Offloading to unreliable edge servers causes failures, which postpone or prolong task offloading. The problem becomes severe when offloading latency-sensitive mobile applications (e.g., mobile augmented reality applications, traffic safety), where timing deadline violations can lead to performance degradation or even life-threatening situations. Estimating reliability is non-trivial and complex due to the resource heterogeneity of edge servers, the constant mobility of user mobile devices and their exposure to different edge servers during mobility, and dynamic and volatile application workloads that occur on edge servers due to concurrent task offloading from multiple mobile devices.

Solution Technical Requirements

To address and cope with reliability challenges in edge offloading, we devised requirements for a reliable solution. The solution itself must (i) be adaptable to dynamic changes in the edge environment to address failures and heterogeneous resources, (ii) track long-term performance to identify reliable edge servers and quantify their reliability, and (iii) ensure that reliability information is shared and accessible globally to all mobile devices without experiencing data loss or tampering, where malicious actors could manipulate reliability information to their advantage (e.g., presenting unreliable servers as reliable ones and causing offloading failures at the cost of the user and service provider). These requirements are important during mobility between different cell sites, where a mobile device encounters edge servers with whom it has had no prior experience and cannot judge their reliability level. However, sharing and accessing reliability information assessed by other mobile devices that had prior interaction with target edge servers can be critically valuable for informative and reliable offloading decision-making.

Blockchain-based Reputation System as Proposed Solution for Reliability

The solution that best satisfies all three requirements is a blockchain-based reputation system. The distributed reputation system collects, aggregates, and assesses targets' behavior based on past actions and results. In our context, past behavior refers to the historical performance of edge servers when executing offloaded tasks. When failures occur during offloading, they directly impact performance (e.g., longer application response times), which can also be measured by the distributed reputation system. Hence, the first two requirements of dynamic adaptability and long-term performance tracking are accommodated. To address the third requirement, we employ blockchain. Blockchain is a decentralized peer-to-peer network that secures transactions through consensus, a protocol where all participating blockchain nodes agree on the current blockchain state. The blockchain state consists of all transactions recorded on the blockchain, representing events of data change, performed computations, network transfers, and similar activities. It is difficult to tamper with transactions without compromising the majority of nodes, and it is extremely cumbersome on a large-scale public blockchain due to its size (e.g., Ethereum). Therefore, it ensures protection against data tampering, prevents data loss, and is accessible globally. Storing reliability information on the public blockchain satisfies the third requirement. To meet all three requirements at once, we integrate blockchain and reputation systems to achieve dynamic adaptability, long-term performance tracking, and secure global storage and accessibility of reliability information.

Emerged Slow Consensus Challenge

Despite significant promises in addressing reliability issues in edge offloading, the blockchain-based reputation system has an issue with the slow consensus protocol, which becomes more time-consuming as the blockchain network size increases. A time-consuming consensus contradicts our objective of providing low-latency task offloading, which is critical for latency-sensitive mobile applications. We must strike a balance or trade-off between blockchain response time and its secure and accessible global reliability information. We will address this issue in our next blog and evaluate our overall solution, which was presented in all previous blog entries.