International Business Machines Corporation

Keqiang He

Kanak Behari Agarwal

Eric Rozner

Wesley Michael Felter

Yu Gu

John Bruce Carter

At an application executing in conjunction with a vSwitch in a host system, using a processor assigned to the vSwitch in the host system, a flow of a number of packets is received from a VM. At the application, a set of CWND values is computed using a corresponding set of congestion control algorithms. At the application, a determination is made whether any of the CWND values in the set of CWND values match the number of packets in the flow within a tolerance value. In response to a CWND value in the set of CWND matching the number of packets in the flow within the tolerance value, a conclusion is drawn that a type of the congestion control algorithm which computed the matching CWND value is the type of a local congestion control algorithm implemented within the VM.

Identifying a local congestion control algorithm of a virtual machine

At an application executing in conjunction with a vSwitch a determination is made that a first flow from a first VM is experiencing congestion. The first flow is selected for throttling. a second flow is also selected for throttling, the second flow using a portion of a network path used by the first flow in a data network. At the application, a total CWND adjustment is distributed between the first flow and the second flow. A first CWND value associated with the first flow is adjusted by a first portion of the total CWND window, and a second CWND value associated with the second flow is adjusted by a second portion of the total CWND window.

Virtual switch-based congestion control for multiple TCP flows

A timer is associated with a packet of a flow from a VM at an application executing in conjunction with a vSwitch in a host system, using a processor assigned to the vSwitch in the host system. At the application, using a counter, a number of packets of the flow that are received and acknowledged in response packets is counted, the response packets being received from a receiver of the flow. At the application, using a period measured by the timer and the number of received packets acknowledged as counted by the counter, a CWND value is computed. The CWND value is applied to the flow at the vSwitch such that the vSwitch transmits, from the flow to a network, only a number of packets up to the CWND value.

Virtual switch-based congestion control for datacenter networks

At an application executing in conjunction with a vSwitch in a host system, a CWND value is computed corresponding to a flow from a VM using a period measured by a timer and a number of packets of the flow received and acknowledged in response packets, the number being counted by a counter, the timer being associated with a packet of the flow. The CWND value is stored in a field in a response packet received from a receiver of the flow, the field being designated for carrying a RWND value, the response packet corresponding to a packet in the flow. The storing forms a modified response packet.

Enforcing datacenter-level congestion control

At an application executing in conjunction with a vSwitch in a host system, using a processor assigned to the vSwitch in the host system, a CWND value corresponding to a flow from a VM is computed using a period measured by a timer and a number of packets of the flow received and acknowledged in response packets, the number being counted by a counter, the timer being associated with a packet of the flow. A set of flow parameters is extracted, at the application, from the flow. At the application, a normalized value corresponding to the flow is computed. At the application, the CWND value is reduced according to the normalized value of the flow to form a reduced CWND value. The reduced CWND value is applied to the flow at the vSwitch such that the vSwitch transmits, from the flow to a network, only a number of packets up to the reduced CWND value.

Normalized flow value-based congestion control

At an application executing in conjunction with a vSwitch in a host system, using a processor assigned to the vSwitch in the host system, a CWND value is computed corresponding to a flow from a VM using a period measured by a timer and a number of packets of the flow received and acknowledged in response packets, the number being counted by a counter, the timer being associated with a packet of the flow. The CWND value is stored in a field in a response packet received from a receiver of the flow, the field being designated for carrying a RWND value, the response packet corresponding to a packet in the flow. The storing forms a modified response packet. The modified response packet is sent from the vSwitch to the VM.

A method, system, and computer program product for determining a congestion value in a networking component in a data network are provided in the illustrative embodiments. A first packet is set to use a first priority level and a second packet is set to use a second priority level. The second priority level is lower than the first priority level. The second packet is transmitted after the first packet to the networking component, causing the networking component to transmit the first and the second packets after they are received at the networking component. A function is applied to an arrival delay to compute the congestion value of the networking component, wherein the arrival delay comprises a length of an elapsed period between receiving a transmission of the first packet from the networking component and receiving a transmission of the second packet from the networking component.

Congestion detection using a differentiated packet pair

A network adapter for handling packet reordering receives a set of packets. The network adapter is configured to aggregate a subset of in-sequence packets from the set of packets into at least one segment before passing the at least one segment from the network adapter to a next layer in a protocol stack. An in-sequence packet from the set of packets is stored into a first segment. Responsive to receiving an out of sequence packet in the set of packets, the out of sequence packet is stored in a second segment. An adaptive timeout timer is activated responsive to receiving the out of sequence packet, and the first segment is held at the network adapter while the adaptive timeout timer runs. Prior to the adaptive timeout timer expiring, a missing packet is received and included in the first segment. The network adapter sends the first segment to the next layer.

Growing community of inventors

Madison, WI, United States of America

Keqiang He