DECO: Joint Computation Scheduling, Caching, and Communication in Data-Intensive Computing Networks

Rate Allocation and Content Placement in Cache Networks

“…the number of variables, for the loose and tight settings, respectively. We observe that in the loose setting the execution times for LBSB are comparable with greedy algorithms; however, in the tight setting LBSB is much slower, as the the trust-region algorithm that we use at each iteration of LBSB requires more iterations to satisfy(15). Fig.3cshows the objective and feasibility trajectories for the iterations of LBSB; as iterations progress, feasibility improves and thusly the objective value decreases.…”

“…The same problem was also studied, albeit in a different model, by Dehghan et al [28]. Online cache allocation algorithms exist, e.g., for maximizing throughput [15], [29], or minimizing delay [30]. Ioannidis and Yeh [9] study a similar problem as Shanmugam et al [25] for networks with arbitrary topology and linear link costs, seeking cache allocations that minimize routing costs across multiple hops.…”

“…If the traffic rate on a link is greater than the link capacity, the network becomes unstable. In order for (6) to ensure stability, similar to [15], [29], in effect we assume that the size of requests are negligible compared to the size of requested items, and the load primarily consists of the downstream traffic of items. Note that the load on edge (b, a) depends both the rate and the cache allocation strategy, while constraints (6) are non-convex.…”

“…3b), however, we see that the execution time of LBSB is higher, particularly when the number of variables is large (corresponding to larger topologies, i.e., grid-2d, balanced-tree, hypercube, small-world, and erdos-renyi). The main reason is that, in the tight setting, the trust-region algorithm used as a subroutine at each iteration requires a higher number of iterations to satisfy (15); as a result, the execution time for LBSB increases. Nonetheless, as we observed in Fig.…”

“…Instead, they are entities capable of storing data, performing computations, and making decisions. Nodes can thus fetch user-requested content [13], [14], or perform user-specified computation tasks [15], [16], instead of simply maintaining point to point communication sessions. In turn, such functionalities can address the ever increasing interest in running data-intensive applications in large-scale networks, such as machine learning at the edge [17], IoT-enabled health care [18], and scientific data-intensive computation [19], [20].…”