News Column

Patent Application Titled "Multi-Stage Large Send Offload" Published Online

July 3, 2014

By a News Reporter-Staff News Editor at Computer Weekly News -- According to news reporting originating from Washington, D.C., by VerticalNews journalists, a patent application by the inventors Starks, John A. (Seattle, WA); Mange, Keith L. (Bellevue, WA), filed on February 17, 2014, was made available online on June 19, 2014.

The assignee for this patent application is Microsoft Corporation.

Reporters obtained the following quote from the background information supplied by the inventors: "When a guest computer system is emulated on a host computer system, the guest computer system is called a 'virtual machine' as the guest computer system only exists in the host computer system as a software representation of the operation of one specific hardware configuration that may diverge from the native machine. The virtual machine presents to the software operating on the virtual machine an emulated hardware configuration.

"A virtual machine management system (sometimes referred to as a virtual machine monitor or a hypervisor) is also often employed to manage one or more virtual machines so that multiple virtual machines can run on a single computing device concurrently. The virtual machine management system runs directly on the native hardware and virtualizes the resources of the machine by exposing interfaces to virtual machines for access to the underlying hardware. A host operating system and a virtual machine management system may run side-by-side on the same physical hardware. For purposes of clarity will we use the abbreviation VMM to refer to all incarnations of a virtual machine management system.

"One problem that occurs in the operating system virtualization context relates to computing resources such as data storage devices, data input and output devices, networking devices etc. Because each of host computing device's multiple operating systems may have different functionality, there is a question as to which computing resources should be apportioned to which of the multiple operating systems. For example, a virtualized host computing device may include only a single network interface card (NIC) that enables the host computing device to communicate with other networked computers. This scenario raises the question of which of the multiple operating systems on the virtualized host should be permitted to interact with and control the NIC.

"When one of the operating systems controls the NIC, the other operating systems sends it packets to the network through the operating system that controls the NIC. In such a case, the packet size accepted by the NIC may not be known. However, sending network TCP packets through a network stack is computationally expensive. Resources must be allocated for each packet, and each component in the networking stack typically examines each packet. This problem is compounded in a virtualization environment, because each packet is also transferred between the guest VM to the root operating system. This entails a fixed overhead per packet that can be quite large. On the other hand, the networking stack packet size is normally limited by the maximum transmission unit (MTU) size of the connection, e.g, 1500 bytes. It is not typically feasible to increase the MTU size since it is limited by network infrastructure.

"Hardware NICs provide a feature called 'Large Send Offload' (LSO) that allows larger TCP packets to travel through the stack all the way to the NIC. Since most of the cost per packet is fixed, this does a fairly good job, but NICs typically support packets that are fairly small, around 62 KB. There is a need for the transmission between operating systems of larger packets to reduce overhead."

In addition to obtaining background information on this patent application, VerticalNews editors also obtained the inventors' summary information for this patent application: "The embodiments described allow a network stack to send very large packets, larger than a physical NIC typically supports with large segment offload (LSO). In general, this is accomplished by performing multi-pass LSO. A first-stage LSO filter is inserted somewhere between the network stack and the physical NIC. The first-stage filter splits very large LSO packets into LSO packets that are small enough for the NIC. The NIC then performs a second pass of LSO by splitting these sub-packets into standard MTU-sized networking packets for transmission on the network.

"To that end, a first operating system operating on a computing device receives an indicator of a first LSO packet size. The first LSO packet size is a multiple of a second LSO packet size that is supported by a network interface card connected to the computing device. The first operating system formats data (e.g., from an application) into a first packet of a first LSO packet size. The first packet is then transferred to a second operating system on the same computing device that has access to a network interface card. The first packet is then split on the second operating system into multiple LSO packets of a second LSO packet size that can be consumed by the network interface card. The multiple LSO packets are sent to the network interface card for transmission on the network in packets of a size supported by the network.

"In general, the first operating system is executing on a virtual machine and the indicator of a first LSO packet size is received from a hypervisor operating on the same computing device. The virtual machine can be migrated to a second computing device and another indicator of a first LSO packet size is received from a hypervisor operating on the second computing device. The indicator of the first LSO packet size received from the hypervisor operating on the second computing device can different from the indicator of the first LSO packet size received from the hypervisor on the computing device. Consequently, the indicator of the first LSO size received from each of the hypervisor operating on the computing device and the hypervisor operating on the second computing device can be tuned for the specific computing device's CPU usage, throughput, latency or any combination thereof

"In general, the first packet has a TCP header. The packet header from the first packet is copied to the packets of second LSO-sized packets when they are split out.


"The foregoing summary, as well as the following detailed description of preferred embodiments, is better understood when read in conjunction with the appended drawings. For the purpose of illustrating the invention, there is shown in the drawings exemplary constructions of the invention; however, the invention is not limited to the specific methods and instrumentalities disclosed. In the drawings:

"FIG. 1 is a block diagram representing a computer system in which aspects of the present invention may be incorporated;

"FIG. 2 illustrates a virtualized computing system environment;

"FIG. 3 illustrates the communication of networking across a virtualization boundary;

"FIG. 4 is a flow diagram of the packet processing in accordance with an aspect of the invention; and

"FIG. 5 is a flow diagram of the processing performed by the virtual switch according to an aspect of the invention."

For more information, see this patent application: Starks, John A.; Mange, Keith L. Multi-Stage Large Send Offload. Filed February 17, 2014 and posted June 19, 2014. Patent URL:

Keywords for this news article include: Software, Microsoft Corporation.

Our reports deliver fact-based news of research and discoveries from around the world. Copyright 2014, NewsRx LLC

For more stories covering the world of technology, please see HispanicBusiness' Tech Channel

Source: Computer Weekly News

Story Tools Facebook Linkedin Twitter RSS Feed Email Alerts & Newsletters