Nvidia Corporation

Ashutosh Jha

Anand Rajagopalan

Sameer Nanda

Ayaz Abdulla

Thomas A Maufer

Hemamalini Manickavasagam

Norman K Chen

Paul J Sidenblad

Radoslav Danilak

Paul Gyugyi

Srinivas Ramachandran

David Lim

Hemamalinl Manickavasagam

Paul J Guygyi

Ajay Kumar Gupta

Venkat Tammineedi

In various examples, access to VM memory by virtualization software is secured using a trusted firmware of a host controller to validate one or more of a command to read a VM's memory and/or the data read from VM memory in order to protect against improper access to data in VM memory. If validation fails, the firmware may refrain from reading the data and/or from providing the virtualization software with access to the data. The data may include a request command from a VM regarding establishing or modifying a connection using the host controller to another entity, such as another device within or outside of the virtualization environment. The virtualization software may use the request command to facilitate the connection. The host controller may provide an eXtensible Host Controller Interface (xHCI) or a different type of interface for the connection.

Secure access of virtual machine memory suitable for AI assisted automotive applications

A user datagram protocol (UDP) is a well-known protocol for transferring data between two nodes of a network. When data is too large to fit within a single UDP packet that can be transmitted between the two nodes, the data needs to be segmented and transmitted with multiple packets and reassembled on the receiving node. Techniques are disclosed herein, for example, for offloading such segmentation, transmission, and reassembly from the central processing units (CPUs) of the nodes. Such offloading is performed efficiently, for example, by repurposing legacy protocol fields used in UDP transmission such as the internet protocol (IP) identification (ID), time to live (TTL), type of service (TOS), and/or EtherType fields to encode information needed for efficient segmentation, out of order reception, and reassembly.

Techniques for enhancing the UDP network protocol to efficiently transmit large data units

A method for identifying a failed network interface card in a system having two NICs configured as a team includes the steps of transmitting a first data packet from the first NIC to a third NIC, wherein the third NIC is not a member of the team, and transmitting a second data packet from the first NIC to the second NIC or from the second NIC to the third NIC, depending on whether the third NIC responds to the transmission of the first data packet. One advantage of the disclosed method is that it specifically identifies which NIC within the team has failed, which is something that cannot be determined by simply exchanging packets between the two NICs.

Technique for identifying a failed network interface card within a team of network interface cards

A method and system for dynamically controlling scaling in a computing device is disclosed. Specifically, in one embodiment, the system information of the computing device is collected and is compared with a trigger condition to generate a comparison result. According to the comparison result, the distribution of a processing task to handle network traffic received by the computing device to at least one designated processing unit in this computing device is either enabled or disabled.

Dynamic control of scaling in computing devices

The invention sets forth an approach for aggregating a plurality of NICs in a computing device into a single logical NIC as seen by that computing device's operating system. The combination of the single logical NIC and a network resource manager provides a reliable and persistent interface to the operating system and to the network hardware, thereby improving the reliability and ease-of-configuration of the computing device. The invention also may improve communications security by supporting the 802.1X and the 802.1Q networking standards.

Single logical network interface for advanced load balancing and fail-over functionality

One embodiment of the present invention sets forth a method for failing back network connections to a network interface card (NIC) within a computing device. The method includes the steps of monitoring a failed or unreliable NIC within the computing device, determining that the failed or unreliable NIC has recovered, determining that a functional NIC within the computing device is overloaded, selecting a first connection set communicating through the overloaded NIC, and transferring the first connection set to the recovered NIC. With this approach, intelligent decisions can be advantageously made regarding whether to fail back a network connection set to a recovered NIC based on the traffic loads on the overloaded NIC and the recovered NIC. Such an approach to balancing network traffic across the functional NICs within a computing device may substantially improve overall performance relative to prior art techniques.

Intelligent failback in a load-balanced networking environment

A system and methods of uploading payload data to user buffers in system memory and of uploading partially processed frame data to legacy buffers allocated in Operating System memory space are described. User buffers are stored in a portion of system memory allocated to an application program, therefore data stored in user buffers does not need to be copied from another portion of system memory to the portion of system memory allocated to the application program. When partially processed frame data is uploaded by hardware to a legacy buffer in system memory, a tag, uniquely identifying the legacy buffer location is transferred by the hardware to a TCP stack, enabling the TCP stack to locate the legacy buffer.

Uploading TCP frame data to user buffers and buffers in system memory

A hash engine in a network device driver maintains data on the utilization and error rate for each network interface card ('NIC') within a local computing device. From this data, the hash engine intelligently selects transmit NICs and receive NICs based on various networking parameters provided from a software driver program. Transmit packets sent from the operating system in a local computing device to a remote computing device are intercepted, modified and redirected to transmit NICs selected by the hash engine for transmission to remote computing devices. Similarly, address resolution protocol ('ARP') response packets sent by the operating system in response to ARP request packets are intercepted, modified and redirected to receive NICs selected by the hash engine for transmission. By selecting receive NICs and transmit NICs in this fashion, the hash engine is able to intelligently load balance transmit and receive traffic in the local computing device, thereby improving overall network performance relative to prior art techniques.

Intelligent load balancing and failover of network traffic

A method for establishing a network connection between two computing devices within the same computer network includes the steps of generating a masquerade IP address request, where the masquerade IP address request includes a masquerade MAC address, transmitting the masquerade IP address request to a DHCP server, and receiving a masquerade IP address from the DHCP server. The masquerade IP address is then used as the sender's IP address in an ARP broadcast request transmitted to set up the network connection. Since the masquerade IP address is unique relative to the computer network, computing devices within the network do no overwrite existing IP-to-MAC relationships in their respective ARP caches with the IP-to-MAC relationship reflected in the ARP broadcast request. Thus, the method enables a network connection to be initiated between two computing devices in the same computer network while avoiding ARP cache pollution on other computing devices in that network.

System and method for avoiding ARP cache pollution

A method of avoiding cache corruption when establishing a network connection includes the steps of transmitting a request to a computing device, where the request includes a masquerade layer-3 address, and receiving a reply transmitted by the computing device in response to the request, where the reply includes a MAC address associated with the computing device. Since the masquerade layer-3 address is unique relative to the computer network, computing devices within the network do no overwrite existing layer-3-to-MAC relationships in their respective caches with the layer-3-to-MAC relationship reflected in the request. Thus, the method enables a network connection to be initiated between two computing devices in the same computer network while avoiding neighbor cache pollution on other computing devices in that network.

System and method for avoiding neighbor cache pollution

A method and apparatus for transmitting commands between a TCP stack and an offload unit and for communicating receive and transmit data buffer locations is described. A command ring buffer stored in system memory is used to transmit commands from the TCP stack to the offload unit and to transmit command status from the offload unit to the TCP stack. A notification ring buffer is used to transmit connection information from the offload unit to the TCP stack. Other ring buffers are used to transmit locations of transmit buffers or receive buffers stored in system memory from the TCP stack to the offload unit.

Transmitting commands and information between a TCP/IP stack and an offload unit

One embodiment of the invention is an architecture for improving the performance of a computer system containing a plurality of hardware input/output devices. The architecture implements an operating system configured to perform all related input/output operations within the operating system kernel. Thus, the operating system enables a first device driver that produces data to pass data directly to a second device driver that consumes data, without a context switch. One advantage of this approach is that computer system performance may be substantially increased due to a reduction in context switching.

Growing community of inventors

Sunnyvale, CA, United States of America

Ashutosh Jha