Composable architecture for rack scale big data computing

Chung-Sheng Li, Hubertus Franke, Colin Parris, Bulent Abali, Mukil Kesavan, Victor Chang
2017 Future generations computer systems  
Keywords: Big data platforms, Composable system architecture, Disaggregated datacenter architecture, composable datacenter, software defined environments, software defined networking. Abstract: The rapid growth of cloud computing, both in terms of the spectrum and volume of cloud workloads, necessitate re-visiting the traditional rack-mountable servers based datacenter design. Next generation datacenters need to offer enhanced support for: (i) fast changing system configuration requirements due
more » ... to workload constraints, (ii) timely adoption of emerging hardware technologies, and (iii) maximal sharing of systems and subsystems in order to lower costs. Disaggregated datacenters, constructed as a collection of individual resources such as CPU, memory, disks etc., and composed into workload execution units on demand, are an interesting new trend that can address the above challenges. In this paper, we demonstrated the feasibility of composable systems through building a rack scale composable system prototype using PCIe switch. Through empirical approaches, we develop assessment of the opportunities and challenges for leveraging the composable architecture for rack scale cloud datacenters with a focus on big data and NoSQL workloads. In particular, we compare and contrast the programming models that can be used to access the composable resources, and developed the implications for the network and resource provisioning and management for rack scale architecture. throughput and latency are usually poor, due to a prolonged execution time and constrained quality of service (QoS). Disaggregated datacenter, constructed as a collection of individual resources such as CPU, memory, HDDs etc., and composed into workload execution units on demand, is an interesting new trend that satisfies several of the above requirements [9] . In this paper, we demonstrated the feasibility of composable systems through building a rack scale composable system prototype using PCIe switch. Through empirical approaches, we develop assessment of the opportunities and challenges for leveraging the composable architecture for rack scale cloud datacenters with a focus on big data and NoSQL workloads. We compare and contrast the programming models that can be used to access these composable resources. We also develop the implications and requirements for network and resource provisioning and management. Based on this qualitative assessment and early experimental results, we conclude that a composable rack scale architecture with appropriate programming models and resource provisioning is likely to achieve improved datacenter operating efficiency. This architecture is particularly suitable for heterogeneous and fast evolving workload environments as these environments often have dynamic resource requirements and can benefit from the improved elasticity of the physical resource pooling offered by the composable rack scale architecture. The rest of the paper is organized as follows: Section 2 describes the architecture of composable systems for a refactored datacenter. Related work in this area is reviewed in Seciton 3. The software stack for such composable systems is described in Section 4. The network considerations for such composable systems are described in Section 5. A rack scale composable prototype system based on PCIe switch is described in Section 6. We describe the rack scale composable memory in Section 7. Section 8 describes the methodology for distributed resource scheduling. Empirical results from various big data workloads on such systems are reported and discussed in Section 9. Discussions of the implications are summarized in Section 10. COMPOSABLE SYSTEM ARCHITECTURE Composable datacenter architecture, which refactors datacenter into physical resource pools (in terms of compute, memory, I/O, and networking), offers the potential advantage of enabling continuous peak workload performance while minimizing resource fragmentation for fast evolving heterogeneous workloads. Figure 3 shows rack scale composability, which leverages the fast progress of the networking capabilities, software defined environments, and the increasing demand for high utilization of computing resources in order to achieve maximal efficiency. On the networking front, the emerging trend is to utilize a high throughput low latency network as the "backplane" of the system. Such a system can vary from rack, cluster of
doi:10.1016/j.future.2016.07.014 fatcat:bf7rs7plxrb5beczlbljmze3dq