International Business Machines Corporation

Globalfoundries Inc.

Bernard Metzler

Animesh Kumar Trivedi

Patrick M Stuedi

Claude Basso

Philippe Damon

Brahmanand Kumar Gorti

Laurent Frelechoux

Jonas Pfefferle

Philip W Frey

Fredy Daniel Neeser

Bay Van Nguyen

Renato John Recio

Nikolas Ioannou

James Allan Kahle

Colin Beaton Verrilli

Clark Debs Jeffries

Martin L Schmatz

Michele M Franceschini

Natarajan Vaidhyanathan

Blake G Fitch

Andreas Kind

Patrick Droz

James C Sexton

Eran Gampel

Gal Sagi

Ravinder Reddy Amanaganti

Richard John Blasiak

Hoyt Edwin Bowen, Jr

Lars Schneidenbach

Bay V Nguygen

Lakas Kencl

Embodiments for providing enhanced endpoint multicast emulation in a computing environment. One or more multicast operations may be executed on an overlay network using endpoint multicast emulation by using an overlay layer or a virtual extensible LAN ('VXLAN') layer to maintain control over one or more multicast groups.

Enhanced endpoint multicast emulation

An RDMA-capable network interface controller provides an RDMA access to a physical memory using multiple mapping tables; the physical memory includes a plurality of physical memory regions, at least some of which are associated with a virtual memory region. A mapping unit is configured to map memory region identifiers, each of which is adapted to identify a virtual memory region and an associated physical memory region, to virtual memory regions and to the associated physical memory regions based on a mapping table selected from multiple mapping tables based on a network identifier. Each of the mapping tables is indexed using a plurality of memory region identifiers, each associated with a virtual memory region and a physical memory region. A processing unit is configured to receive an access request from a client for accessing one of the physical memory regions associated with a virtual memory region.

Controller and method for migrating RDMA memory mappings of a virtual machine

Embodiments of the present invention provide systems and methods for extending the remote direct memory access (RDMA) operations for accessing data from storage class memory (SCM). The method includes receiving an RDMA request in a first semantic, to a memory in a second semantic. The RDMA request in the first semantic is encoded, by encoding a type of the RDMA request into a memory area identifier, which includes an 8-bit key used to define additional storage semantics for the RDMA operation.

Extending remote direct memory access operations for storage class memory access

A persistent caching system is provided. The persistent caching system includes a storage system having a caching server for storing data, and a client for accessing the data through a network. The caching server is configured to store the data in a number of virtual memory blocks. The virtual memory blocks refer to an associated memory-mapped file in a file system of the caching server. The caching server is configured to export addresses of the virtual memory blocks to the client. The client is configured to access at least some of the virtual memory blocks through RDMA using the exported addresses. The caching server is configured to page virtual memory blocks being accessed by one or more clients through RDMA to and/or from the memory-mapped files associated with the accessed virtual memory blocks.

Persistent caching for operating a persistent caching system

Method and device for data transmissions using RDMA. The present invention is implemented between a first entity using a first data structure type and a second entity using a second data structure type over a third entity. The third entity is coupled to a table caching fingerprints of first data structures of the first data structure type and second data structures of the second data structure type associated therewith. A certain first data structure and the second data structure associated therewith represent a certain, identical RDMA function call. A first data structure representing a certain RDMA function call is sent from the first entity to the third entity; the fingerprint for the sent first data structure is determined; a second data structure associated with the determined fingerprint is looked up in the table; and the looked up second data structure is sent to the second entity.

Method and device for data transmissions using RDMA

A queued, byte addressed system and method for accessing flash memory and other non-volatile storage class memory, and potentially other types of non-volatile memory (NVM) storage systems. In a host device, e.g., a standalone or networked computer, having attached NVM device storage integrated into a switching fabric wherein the NVM device appears as an industry standard OFED&#x2122; RDMA verbs provider. The verbs provider enables communicating with a 'local storage peer' using the existing OpenFabrics RDMA host functionality. User applications issue RDMA Read/Write directives to the 'local peer (seen as a persistent storage) in NVM enabling NVM memory access at byte granularity. The queued, byte addressed system and method provides for Zero copy NVM access. The methods enables operations that establish application private Queue Pairs to provide asynchronous NVM memory access operations at byte level granularity.

Local direct storage class memory access

Embodiments relate to providing remote procedure calls (RPC) with call-by-reference semantics using remote direct memory access (RDMA). An aspect includes establishing a session between a client and a server. A client session heap memory and a server session heap memory are initialized with a RDMA provider. A parameter is then stored in the session heap memory. A handle for the parameter is passed to the server which permits access to the parameter responsive to an on-demand request from the server.

Remote procedure call with call-by-reference semantics using remote direct memory access

A computer-implemented method, system, and article of manufacture for data communication between a requester and a responder in a remote direct memory access (RDMA) network, where each of the requester and the responder is an RDMA-enabled host of the network. The method includes: sending a request for the responder to provide data, where the request includes a mapped steering tag that is obtained by mapping a set of memory buffers of the requester onto a single representation that allows for identifying each of the memory buffers of the set; and receiving the requested data together with the mapped steering tag and assigning the data being received to the memory buffers of the set consistently with the mapping.

RDMA read destination buffers mapped onto a single representation

Exemplary embodiments include RDMA methods and systems for sending application data to a computer memory destination in a direct but non-blocking fashion. The method can include posting a new work request for an RDMA connection or association, determining if there is a prior work request for the same connection or association enqueued for processing, in response to a determination that no prior work request is enqueued for processing, processing the new work request directly by sending RDMA frames containing application data referred to by the work request to the computer memory destination, performing direct sending while there is sufficient send space to process the new work request, and delegating the new work request to asynchronous transmission if a prior work request is already enqueued for processing or lack of send space would block a subsequent transmission operation.

Direct sending and asynchronous transmission for RDMA software implementations

A method for data transmission on a device without intermediate buffering is provided. An application request is received to transmit data from the device to a second device over a network. The data from application memory is formatted for transmitting to the second device. The data are transmitted from the device to the second device without intermediate buffering. A send state is retrieved. The send state is compared to expected send state. If the send state meets the expected send state, a completion of the data transmit request is generated.

Zero copy data transmission in a software based RDMA network stack

A method, computer program product and system for processing TCP/IP packets. A TCP protocol stack may store a payload of a received TCP/IP packet in a data fragment list. The TCP protocol stack may further read the header of the received packet to extract a value used to index into a table storing a list of transport control blocks (TCBs). The TCP protocol stack may further perform a lock and a read operation on the TCB indexed in the table. The TCP protocol stack may further transmit the payload to the TCP application without requiring the application to perform a lock, read, write or unlock operation on the indexed TCB since the TCP protocol stack and the TCP application are operating on the same thread. By the TCP application foregoing the lock, read, write and unlock operations on the TCB, there is a reduction in the number of memory accesses.

Reducing memory accesses in processing TCP/IP packets

Load balancing apparatus for a data communications network comprises hash logic for computing a hash function on incoming data packets. A threshold detector is connected to the hash logic for triggering, in response to utilization of the downstream objects exceeding a predefined threshold, redefinition in the hash logic of parameters of the hash function from a first set of parameters to a second set of parameters for redistributing the data packets amongst the downstream objects. In use, the hash logic, directs the packets for routing to downstream objects in the network via a first routing path based on a hash computation using the first set of parameters, and, if the threshold is exceeded, selectively directs the packets to one of the first routing path and a second routing path in dependence on separate hash computations using the first and the second sets of parameters for subsequent routing of the packets via the selected one of the first and second routing paths based on the results of one of the separate hash computations.

Growing community of inventors

Zurich, Switzerland

Bernard Metzler