Fine-grained quality of service in datacenters through end-host control of traffic flow

A system for providing Quality of Service to traffic flows comprising:

• a computer having a Central Processing Unit (CPU), memory, and a Network Interface Controller (NIC) and hosting a traffic engineering controller executable by the CPU having a global view, provided by the multiple flow agents at multiple end hosts, of multiple long-lived traffic flows in a datacenter and further comprising:
• an interpreter module that receives a report on a long-lived traffic flow identified by a flow agent at an end host, sent from the end host over a network of the datacenter serviced by the traffic engineering controller, and identify the Quality of Service (QoS) from the report;
• and
• an allocation module that:
• assigns an allocation of bandwidth to the long-lived traffic flow that controls a rate at which the long-lived traffic flow is transmitted at the end host, supports the QoS, and reduced bandwidth occupied by the long-term traffic flow at a downstream bandwidth resource shared with a second long-lived traffic flow,
• communicates the allocation of bandwidth to the flow agent to control, from the end host, bandwidth occupied by the long-lived traffic flow, along a path traversed by the long-lived traffic flow, by the flow agent controlling the rate at which the long-lived traffic flow is sent,
• uses the global view of the multiple long-lived traffic flows through the datacenter that share the downstream bandwidth resource to assign multiple allocations of bandwidth to the multiple long-lived traffic flows to prevent overtaxing g the downstream bandwidth resource;
• and
• communicates the multiple allocations of bandwidth to the multiple flow agents at the multiple end hosts originating the multiple long-lived traffic flows to control the multiple long-lived traffic flows from the multiple end hosts without providing additional information to intermediate nodes for control operations;
• wherein the allocation module further applies an augmenting function at least once to improve utilization of available bandwidth in the datacenter, the augmenting function;
• checks for available bandwidth for active flow paths traversed by long-lived traffic flows in the datacenter;
• and
• reallocates additional bandwidth from the available bandwidth to a selected long-lived traffic flow according to a priority determined for the selected long-lived traffic flow by the priority module by increasing a corresponding allocation of bandwidth communicated to a corresponding flow agent and controlling an increased rate at which the selected long-lived traffic flow is sent from a corresponding end host.

A system comprising:

• a computer having a Central Processing Unit (CPU), memory, and hosting a traffic engineering controller with a global view of multiple long-lived traffic flows comprising:
• an interpreter module using reports on the multiple long-lived traffic flows, sent by flow agents at originating end hosts, to acquire a global view of a datacenter by identifying the Quality of Service (QoS), the paths for the multiple long-lived traffic flows across a topology of the datacenter, and to establish bandwidths for bandwidth resources traversed by the paths using a topology description;
• and
• an allocation module that:
• calculates, with the global viewer, a rate at which each of the multiple long-lived traffic flows is transmitted by and from a corresponding flow agent and end host to prevent overtaxing shared bandwidth resources in the datacenter without redirecting the multiple long-lived traffic flows from their paths, and
• communicates, from the traffic engineering controller, each of the multiple rates to a corresponding flow agent at a corresponding end host;
• uses the global view of the multiple long-lived traffic flows through the datacenter that share the downstream bandwidth resource to assign multiple allocations of bandwidth to the multiple long-lived traffic flows to prevent overtaxing the downstream bandwidth resource;
• and
• communicates the multiple allocations of bandwidth to the multiple flow agents at the multiple from the multiple end hosts without providing additional information to intermediate nodes for control operations;
• wherein:
• the traffic engineering controller further comprising a topology module within the interpreter module that;
• maintains a description of at least one of a topology and an overlay topology of the datacenter;
• and
• reads the paths for the multiple long-lived traffic flows through the datacenter;
• wherein the allocation module determines the rates for transmitting the multiple long-lived traffic flows that keep the multiple long-lived traffic flows within bandwidth available to be occupied by the multiple long-lived traffic flows at each bandwidth resource along the paths to control the multiple long-lived traffic flows from the originating end hosts as opposed to assigning a control operation to an intermediate network node in the datacenter;
• the topology module further determines an available bandwidth for downstream bandwidth resources in the datacenter;
• and
• the allocation module further distributes multiple rates to the multiple flow agents, at the originating end hosts, mapped to concurrently pass through the downstream bandwidth resources by the topology module, the multiple rates derived by a QoS-based weighting function.

A system supporting a Quality of Service (QoS), comprising:

• multiple end hosts in a datacenter, each of the multiple end hosts having a Central Processing Unit, memory, and a Network Interface Controller;
• a traffic engineering controller with a global view, servicing the datacenter and hosted by a computer and comprising:
• an aggregation module collects reports on multiple traffic flows from multiple end hosts;
• a path module determines fixed paths for the multiple traffic flows across a topology of the datacenter;
• an availability module establishes available bandwidths for bandwidth resources in the datacenter;
• and
• an allocation module assigns rates at which the multiple traffic flows are sent from the multiple end hosts to prevent the multiple traffic flows across the datacenter from overtaxing bandwidth resources shared by the multiple traffic flows while assigning differing rates to individual traffic flows from the multiple traffic flows as a function of corresponding QoS values individually to the multiple traffic flows;
• uses the global view of the multiple traffic flows through the datacenter that share the downstream bandwidth resource to assign multiple allocations of bandwidth to the multiple traffic flows to prevent overtaxing the downstream bandwidth resource;
• and
• communicates the multiple allocations of bandwidth to the multiple flow agents at the multiple end hosts originating the multiple traffic flows to control the multiple traffic flows from the multiple end hosts without providing additional information to intermediate nodes for control operations;
• and
• multiple flow agents, each flow agent of the multiple flow agents runs within user space at an end host of the multiple end hosts, receive the rates assigned by the allocation module, and control transmission of the multiple traffic flows from the multiple end hosts at the rates corresponding to the multiple traffic flows, while maintaining the fixed paths of the multiple traffic flows unaltered;
• wherein the allocation module further applies an augmenting function to maintain high utilization of available bandwidth, the augmenting function:
• checks for available bandwidth based on multiple reports from the aggregation module, the fixed paths determined by the path module, and the available bandwidths established by the availability module;
• and
• reallocates additional bandwidths from the available bandwidths for the bandwidth resources along the fixed paths of the multiple traffic flows according to the QoS values by increasing the rates communicated to the multiple flow agents residing on corresponding end hosts at which the corresponding traffic flows originate.