Storage

XCo: Explicit Coordination for Preventing Congestion in Data Center Ethernet

Free registration required

Executive Summary

Heavy inflow of data is usually managed by accumulating it via multiple storage nodes. Very often, the storage traffic is required to share the network with non-storage traffic. This can trouble Throughput Collapse (TCP) throughput and cause network congestion because of overload of network traffic. This usually causes the storage client to suffer greatly. The one solution that has worked towards this end is the use of commodity communication technologies such as Fibre Channel over Ethernet (FCoE). These technologies are used by cluster-based storage systems to access stored data on a network. Problems also arise when multiple Virtual Machines (VMs) are consolidated on a lesser number of storage nodes. These issues can also cause an overload to the switch buffers. This paper tries to explain that virtualization has the ability to alleviate problems associated with multiple storage nodes in a data center Ethernet. The paper particularly deals with issues faced by network storage. The paper presents the XCo architecture that enables explicit coordination of network traffic among VMs. This architecture works well with the inexpensive and transparent data center Ethernet and provides balance to any switch-level hardware support. The paper undertakes experiments to prove the ability of XCo architecture to solve problems of multiple network traffic. This is done with proof-of-concept implementation and evaluation as the paper provides design, implementation, and evaluation of this architecture prototype.

  • Format: PDF
  • Size: 307.5 KB