Pavilion Data Systems, Inc.

Broadcom Corporation

Mips Technologies, Inc.

Fusion-io, Inc.

Arm Finance Overseas Limited

Avago Technologies General IP (singapore) Pte. Ltd.

Avago Technologies International Sales Pte. Limited

Karagada Ramarao Kishore

Kiron Balkrishna Malwankar

Ariel Hendel

Sundar Kanthadai

Kevin D Kissell

Peter E Kirkpatrick

Vidya Rajagopalan

Maria Ukanwa

Mohan V Kalkunte

Nicholas Ilyadis

Chien-Hsien Wu

Kjeld Svendsen

Georgi Beloev

Gregory Scherer

Radhika Thekkath

Gideon Intrater

Daehwan Daniel Kim

Srinivas Sampath

Era Kasturia Nangia

Meng-Bing Yu

Michael Ni

Xing Yu Jiang

Xing Xu Jiang

A network-displaced direct storage architecture transports storage commands over a network interface. In one implementation, the architecture maps, at hosts, block storage commands to remote direct memory access operations (e.g., over converged Ethernet). The mapped operations are communicated across the network to a network storage appliance. At the network storage appliance, network termination receives the mapped commands, extracts the operation and data, and passes the operation and data to a storage device that implements the operation on a memory.

Network-displaced direct storage

A processing device receives a request from a remote computing device to read data from a virtual storage device. The processing device identifies a physical storage device to be accessed to satisfy the request. The processing device then sends a sub-request to the physical storage device to access a portion of the physical storage device. Responsive to determining that the physical storage device has not returned a response to the sub-request within a latency threshold time associated with the physical storage device, the processing device initiates recovery of the portion of the physical storage device to recompute contents of the portion of the physical storage device. The processing device then returns a response to the request from the remote computing device.

Consistent latency for solid state drives

A first I/O controller of a storage server sends a first command to a first solid state drive (SSD) of the storage server via a first submission queue of the first SSD, wherein the first command is a first read command or a first write command. The first I/O controller receives a first acknowledgement from the first SSD that the first command has been completed via a first completion queue of the first SSD. A second I/O controller of the storage server sends a second command to the first SSD of the storage server via a second submission queue of the first SSD, wherein the second command is a second read command or a second write command. The second I/O controller receives a second acknowledgement from the first SSD that the second command has been completed via a second completion queue of the first SSD.

Parallel data transfer for solid state drives using queue pair subsets

An apparatus includes a first serial port having a first number of lanes usable to form a first multi-lane link to a solid state drive (SSD) and a second serial port having a second number of lanes that is at least twice the first number of lanes. A first subset of the second number of lanes constitutes a first logical serial port and is usable to form a second multi-lane link to a first transport fabric and a second subset of the second number of lanes constitutes a second logical serial port and is usable to form a third multi-lane link to a second transport fabric. The apparatus further includes a controller and a multiplexer to connect the first serial port to one of the first logical serial port or the second logical serial port.

Solid state drive multiplexer

A storage server comprises a plurality of solid state drives (SSDs), a plurality of input/output (I/O) controllers, a switch, and a management controller. A first SSD of the plurality of SSDs comprises a plurality of queue pairs. The management controller is to allocate a first subset of the plurality of queue pairs in the first SSD to a first I/O controller, wherein the first I/O controller is to use the first subset of the plurality of queue pairs to read from and write to the first SSD. The management controller is further to allocate a second subset of the plurality of queue pairs in the first SSD to a second I/O controller, wherein the second I/O controller is to use the second subset of the plurality of queue pairs to read from and write to the first SSD.

Architecture for flash storage server

System and method for performing a high-bandwidth memory copy. Memory transfer instructions allow for copying of data from a first memory location to a second memory location without the use of load and store word instructions thereby achieving a high-bandwidth copy. In one embodiment, the method includes the steps of (1) decoding a destination address from a first memory transfer instruction, (2) storing the destination address in a register in the bus interface unit, (3) decoding a source address from a second memory transfer instruction, and (4) copying the contents of a memory location specified by the source memory address to a memory location specified by the contents of the register. Other methods and a microprocessor system are also presented.

System and method for improving memory transfer

A storage server comprises a plurality of storage devices arranged in a redundant storage array and a plurality of controllers. One of the controllers determines that a storage device in the redundant storage array has failed, determines a first portion of the storage device to be recovered by a first controller and determines a second portion of the storage device to be recovered by a second controller. The first controller recovers the first portion of the storage device and the second controller recovers the second portion of the storage device.

Distributed rebuild of failed storage device

A processing device of a storage server that manages a plurality of SSDs receives to read data. The processing device determines a first SSD of the plurality of SSDs on which the data is stored, and determines that the first SSD is performing a background operation. The processing device determines whether the data is located on a block of the first SSD that will be unavailable during the background operation. Responsive to determining that the data is located on the block of the first SSD, the processing device reconstructs the data based on additional data on one or more remaining SSDs of the plurality of SSDs. The processing device then generates a response to the request that includes the data.

Proactive scheduling of background operations for solid state drives

A processing device of a storage server that manages a plurality of SSDs receives a request to write data. The processing device determines one or more attributes associated with the request, generates a stream tag for the request based on the one or more attributes, and sends a first write command to a first SSD of the plurality of SSDs. The first write command comprises at least a portion of the data and the stream tag that causes the first SSD to write the portion of the data to a first stream block of the first SSD.

Streams across multiple controllers to improve solid state drive performance

A processing device of a storage server receives a write request to write to a first logical address of a virtual storage device that is mapped to a plurality of SSDs. The processing device stores a data block associated with the request in a buffer, stores an entry for the write request in a queue, the entry comprising a reference to the data block in the buffer, sends a write command to an SSD, the write command comprising the data block, and subsequently receives an acknowledgment that the data block has been written to the SSD. The processing device maintains the data block in the buffer until a combined size of a plurality of data blocks in the buffer is equal to or greater than a threshold, the threshold being based on a number of SSDs in the plurality of SSDs and a block size of the plurality of SSDs.

Minimizing read latency for solid state drives

System, method and apparatus for network congestion management and network resource isolation. A high level network usage and device architecture is provided that can satisfy buffering and network bandwidth resource management for data center networks. The congestion management can be defined to bring the reaction point closer to the network ports. In one embodiment, the reaction point is resident in a network interface card (NIC).

System, method and apparatus for network congestion management and network resource isolation

A virtual hint based data cache way prediction scheme, and applications thereof. In an embodiment, a processor retrieves data from a data cache based on a virtual hint value or an alias way prediction value and forwards the data to dependent instructions before a physical address for the data is available. After the physical address is available, the physical address is compared to a physical address tag value for the forwarded data to verify that the forwarded data is the correct data. If the forwarded data is the correct data, a hit signal is generated. If the forwarded data is not the correct data, a miss signal is generated. Any instructions that operate on incorrect data are invalidated and/or replayed.

Data cache virtual hint way prediction, and applications thereof

A universal network interface controller (UNIC) is provided for interfacing a host computer to a switch fabric, a packet network, or both. The UNIC includes encapsulation logic configured to encapsulate a CBP communication for transmission as switch fabric data on the switch fabric. Finally, the UNIC includes transmit logic configured to transmit the encapsulated CBP communication to the remote CBP device using the switch fabric.

Universal network interface controller

Systems and methods for handling virtual machine packets are provided. In some aspects, a method includes receiving, by a network interface controller, an indicator of a packet from a virtual machine. The indicator bypasses a virtual switch coupled to the virtual machine. The method also includes determining, in response to receiving the indicator, whether the packet is designated for accelerated processing. The method also includes providing the indicator to the virtual switch for processing if the packet is determined not to be designated for accelerated processing. The method also includes processing, by the network interface controller, the packet for transmission if the packet is determined to be designated for accelerated processing.

Growing community of inventors

Saratoga, CA, United States of America

Karagada Ramarao Kishore