Microsoft Technology Licensing, LLC

Shachar Raindel

Jitendra Dattatraya Padhye

Anirudh Badam

Brian Don Zill

Ramakrishna Huggahalli

Daehyeok Kim

Behnaz Arzani

Avi William Levy

Mahmoud S El Haddad

Alireza Khosgoftar Monafared

Xinchen Guo

Lihua Yuan

Daniel Max Firestone

Hongqiang Liu

Yibo Zhu

Rohan Gandhi

Wei Bai

Techniques of packet level redundancy in distributed computing systems are disclosed herein. In one example, upon receiving an original packet to be transmitted from a source host to an application executing at a destination host, the source host generates a duplicated packet based on the received original packet. The source host can then encapsulate the original and duplicated packets with first and second outer headers having first and second header values, respectively, and transmitting the original and the duplicated packets from the source host to the destination host via a first network path and a second network path in the computer network, respectively. Then, the transmitted original and duplicated packets can be de-duplicated at the destination host before providing the de-duplicated original and duplicated packets to the application executing at the destination host.

Packet level redundancy in distributed computing systems

Embodiments of the present disclosure includes techniques for cache memory replacement in a processing unit. A first data production operation to store first data to a first cache line of the cache memory is detected at a first time. A retention status of the first cache line is updated to a first retention level as a result of the first data production operation. Protection against displacement of the first data in the first cache line is increased based on the first retention level. A first data consumption operation retrieving the first data from the first cache line is detected at a second time after the first time. The retention status of the first cache line is updated to a second retention level as a result of the first data consumption operation, the second retention level being a lower level of retention than the first retention level.

Cache replacement policy optimization for producer-consumer flows

Techniques are disclosed for identifying faulty links in a virtualized computing environment. Network path latency information is received for one or more network paths in the networked computing environment. Based on the network path latency information, a probable presence of a faulty component is determined. In response to the determination, physical links for a network path associated with the probable faulty component are identified. Information indicative of likely sources of the probable faulty component is received from multiple hosts of the networked computing environment. Based on the identified physical links and information, a faulty component is determined.

Link fault isolation using latencies

Embodiments of the present disclosure includes techniques for partial memory updates in a computer system. A data structure template is received. A first write data of a first write operation is received from a first data source, the first write operation performed in connection with provisioning of a first data payload to memory communicatively coupled with a processing unit. A first merge operation is performed involving the first write data and the first data structure template to obtain a first data structure update. The first data structure update is written to the memory, thereby improving efficiency of updating a first data structure associated with the first data payload.

Partial memory updates using preset templates

In one example of the present technology, an input/output memory management unit (IOMMU) of a computing device is configured to: receive a prefetch message including a virtual address from a central processing unit (CPU) core of a processor of the computing device; perform a page walk on the virtual address through a page table stored in a main memory of the computing device to obtain a prefetched translation of the virtual address to a physical address; and store the prefetched translation of the virtual address to the physical address in a translation lookaside buffer (TLB) of the IOMMU.

Address translation prefetching for input/output devices

Techniques of time-sensitive data delivery in distributed computing systems are disclosed herein. In one example, a server can disseminate the same information to multiple endpoints in a distributed computing system by transmitting multiple packets to the multiple endpoints hosted on additional servers in the distributed computing system. The multiple packets individually include a header field containing a delivery time before which the packets are not forwarded to corresponding final destinations and a payload containing data representing copies of information identical to one another destined to the multiple endpoints hosted on the additional servers.

Time-sensitive data delivery in distributed computing systems

Techniques for network latency estimation in a computer network are disclosed herein. One example technique includes instructing first and second nodes in the computer network to individually perform traceroute operations along a first round-trip route and a second round-trip route between the first and second nodes. The first round-trip route includes an inbound network path of an existing round-trip route between the first and second nodes and an outbound network path that is a reverse of the inbound network path. The second round-trip route has an outbound network path of the existing round-trip route and an inbound network path that is a reverse of the outbound network path. The example technique further includes upon receiving traceroute information from the additional traceroute operations, determine a latency difference between the inbound and outbound network paths of the existing round-trip route based on the received additional traceroute information.

Network latency estimation in distributed computing systems

Techniques for facilitating load balancing in distributed computing systems are disclosed herein. In one embodiment, a method includes receiving, at a destination server, a request packet from a load balancer via the computer network requesting a remote direct memory access ('RDMA') connection between an originating server and one or more other servers selectable by the load balancer. The method can also include configuring, at the destination server, a rule for processing additional packets transmittable to the originating server via the RDMA connection based on the received reply packet. The rule is configured to encapsulate an outgoing packet transmittable to the originating server with an outer header having a destination field containing a network address of the originating server and a source field containing another network address of the destination server.

Load balancing in distributed computing systems

Distributed storage systems, devices, and associated methods of data replication are disclosed herein. In one embodiment, a server in a distributed storage system is configured to write, with an RDMA enabled NIC, a block of data from a memory of the server to a memory at another server via an RDMA network. Upon completion of writing the block of data to the another server, the server can also send metadata representing a memory location and a data size of the written block of data in the memory of the another server via the RDMA network. The sent metadata is to be written into a memory location containing data representing a memory descriptor that is a part of a data structure representing a pre-posted work request configured to write a copy of the block of data from the another server to an additional server via the RDMA network.

Group-based data replication in multi-tenant storage systems

An originating host device in a distribution chain is provided upstream from multiple host devices including intermediary and terminating host devices. The originating host device includes a core with a generation application and a first RDMA NIC. The core: determines a plan for transferring data between the originating host device and the other host devices; and generates WQEs to implement the plan. The first RDMA NIC includes a read application, a descriptor application, and a reception application. The read application is configured such that the first RDMA NIC reads the WQEs from a first memory. The descriptor application is configured such that the first RDMA NIC: writes a portion of the WQEs directly from the first RDMA NIC to a second memory of the intermediary host device; and triggers a second RDMA NIC of the intermediary host device to process the portion of the WQEs stored in the second memory.

Growing community of inventors

Redmond, WA, United States of America

Shachar Raindel