Volley: Automated Information Placement For Geo-Distributed Cloud Services

Datacenters today are distributed across the globe, nonetheless they necessitate to portion information alongside other datacenters equally good equally their clients. This paper from Microsoft Research presents a heuristic strategy for information placement to these geo-distributed datacenters. While at that topographic point has been previous run on information placement inwards LANs together with WSNs, Volley is the outset heuristic for information placement strategies for WANs.

A elementary heuristic is to house each information to the datacenter closest to the customer of that data. But things are non that simple, at that topographic point are several additional constraints to travel considered, including concern constraints, WAN bandwidth costs, datacenter capacity limits, information interdependencies, user-perceived latency, etc. For example, it makes to a greater extent than feel to collocate information that are tightly-coupled/interdependent, such equally ii friends inwards Facebook that update each other walls. As some other example, the frequency of the clients accessing the information needs to travel taken inwards to concern human relationship equally well. As alive mesh together with alive messenger traces show, at that topographic point is meaning information sharing across clients (Figure 5), at that topographic point tin travel meaning benefits to placing information closest to those who role it around heavily, rather than simply placing it simply about some detail customer that accesses the data. Finally, the alive mesh together with messenger traces likewise exhibit that a meaning division of the clients motion together with modify locations (Figure 7), together with thus the ideal placement of information needs to travel changed adaptively equally well.

Algorithm

Volley takes equally input the asking logs for information inwards the system, analyzes them, together with outputs the results on where to best house the data. Volley is non concerned alongside the actual migration, that should travel handled past times other applications.

Volley is an iterative algorithm. In stage 1, Volley computes an initial placement based on customer IPs. In stage 2, Volley iteratively moves information to cut down latency. Finally inwards stage 3, Volley iteratively map information items into datacenters taking into concern human relationship the datacenter capacities. In social club

Evaluation

The alive mesh together with alive messenger traces are used to evaluate Volley via emulations. In these emulations, 12 DCs are assumed. Volley is compared alongside the commonIP protocol (which places information equally closed equally possible to the IP address that around ordinarily accesses it), hashing protocol (which randomly places information to 1 the 12 DCs to optimize for load-balancing), together with oneDataCenter (which places all the information inwards 1 datacenter).

Among these protocolsVolley is the 1 alongside lowest latency, together with hash is unsurprisingly the 1 alongside the highest latency. Hash protocol leads to a lot of inter-datacenter traffic equally it oftentimes places interdependent information inwards dissimilar datacenters. Volley has the to the lowest degree inter-datacenter traffic, of cast excluding the oneDC protocol which evidently has no inter-datacenter traffic.

The evaluations exhibit that Volley converged afterwards a small-scale release of iterations, together with reduced skew past times 2x, inter-datacenter traffic past times 1.8x, together with latency past times 30%. The newspaper does non brand whatever optimality claims for Volley, equally it simply uses heuristics. The existent contribution of Volley is reported equally automating the information placement process.

Share this post