-
公开(公告)号:US09509616B1
公开(公告)日:2016-11-29
申请号:US14552373
申请日:2014-11-24
Applicant: Amazon Technologies, Inc.
Inventor: Alan Michael Judge , Matthew Shawn Wilson
IPC: H04L12/28 , H04L12/803 , H04L12/46 , H04L12/801 , H04L12/813
CPC classification number: H04L47/125 , H04L12/4633 , H04L47/11 , H04L47/20
Abstract: Encapsulated packets may be generated for different packets transmitted between a source instance and destination instance in a computer system. The source instance and destination instance may be implemented by different physical hosts linked by multiple network paths. Congestion of the multiple network paths may be determined and path-balancing polices may be implemented in response to the determined congestion. Each encapsulation packet comprises contents of a corresponding packet, and one or more data values selected in accordance with a path-balancing policy. The data values added to one encapsulation packet may differ from those added to another. Different network paths to the destination may be selected for different encapsulation packets of a given transmission based at least in part on the added data values.
Abstract translation: 可以为在计算机系统中的源实例和目的地实例之间传输的不同数据包生成封装的数据包。 源实例和目的实例可以由通过多个网络路径链接的不同物理主机来实现。 可以确定多个网络路径的拥塞,并且可以响应于所确定的拥塞来实现路径平衡策略。 每个封装分组包括相应分组的内容,以及根据路径平衡策略选择的一个或多个数据值。 添加到一个封装数据包的数据值可能与添加到另一封装数据包的数据值不同。 至少部分地基于所添加的数据值,可以为给定传输的不同封装分组选择到目的地的不同网络路径。
-
公开(公告)号:US11863406B2
公开(公告)日:2024-01-02
申请号:US17466944
申请日:2021-09-03
Applicant: Amazon Technologies, Inc.
Inventor: Robert Michael Johnson , Nafea Bshara , Matthew Shawn Wilson
IPC: H04L12/24 , G06F9/50 , G06F15/78 , H04L41/5054 , H04L41/50
CPC classification number: H04L41/5054 , G06F9/50 , G06F15/7871 , H04L41/5096
Abstract: Methods and apparatus are disclosed for programming reconfigurable logic devices such as FPGAs in a networked server environment. In one example, a system hosting a network service providing field programmable gate array (FPGA) services includes a network service provider configured to receive a request to implement application logic in a plurality of FPGAs, allocate a computing instance comprising the FPGAs in responses to receiving the request, produce configuration information for programming the FPGAs, and send the configuration information to an allocated computing instance. The system further includes a computing host that is allocated by the network service provider as a computing instance which includes memory, processors configured to execute computer-executable instructions stored in the memory, and the programmed FPGAs.
-
公开(公告)号:US20230155957A1
公开(公告)日:2023-05-18
申请号:US18154769
申请日:2023-01-13
Applicant: Amazon Technologies, Inc.
Inventor: Andra-Irina Paraschiv , Matthew Shawn Wilson
IPC: H04L47/78 , G06F9/38 , G06F9/455 , H04L47/762 , H04L47/70
CPC classification number: H04L47/782 , G06F9/3877 , G06F9/45558 , H04L47/762 , H04L47/828 , G06F2009/45562 , G06F2009/45566 , G06F2009/4557 , G06F2009/45595
Abstract: At a first compute instance run on a virtualization host, a local instance scaling manager is launched. The scaling manager determines, based on metrics collected at the host, that a triggering condition for redistributing one or more types of resources of the first compute instance has been met. The scaling manager causes virtualization management components to allocate a subset of the first compute instance's resources to a second compute instance at the host.
-
公开(公告)号:US11558311B2
公开(公告)日:2023-01-17
申请号:US16737780
申请日:2020-01-08
Applicant: Amazon Technologies, Inc.
Inventor: Andra-Irina Paraschiv , Matthew Shawn Wilson
IPC: H04L47/78 , G06F9/38 , G06F9/455 , H04L47/762 , H04L47/70
Abstract: At a first compute instance run on a virtualization host, a local instance scaling manager is launched. The scaling manager determines, based on metrics collected at the host, that a triggering condition for redistributing one or more types of resources of the first compute instance has been met. The scaling manager causes virtualization management components to allocate a subset of the first compute instance's resources to a second compute instance at the host.
-
公开(公告)号:US11042496B1
公开(公告)日:2021-06-22
申请号:US15282487
申请日:2016-09-30
Applicant: Amazon Technologies, Inc.
Inventor: Christopher James BeSerra , Kypros Constantinides , Uwe Dannowski , Nafea Bshara , Matthew Shawn Wilson
Abstract: Provided are systems and methods for enabling peer-to-peer communications between peripheral devices. In various implementations, a computing system can include a PCI switch device. The first PCI switch device can include a first port and be communicatively coupled to a first root complex port. The first PCI switch device can have access to a first PCI endpoint address range. The computing system can further include a second PCI switch device. The second PCI switch device can include a second port, connected to the first port. The second PCI switch device can be communicatively coupled to a second root complex port that is different from the first root complex port. The second PCI switch device can receive a transaction addressed to the first PCI endpoint address range, and identify the transaction as associated with the second port. The second PCI switch device can subsequently transmit the transaction using the second port.
-
公开(公告)号:US10996969B1
公开(公告)日:2021-05-04
申请号:US15825068
申请日:2017-11-28
Applicant: Amazon Technologies, Inc.
Abstract: A server computer toggles between a protected mode and an unprotected mode. In the protected mode, users are unable to access configuration information due to a Base Address Register (BAR) being cleared. However, a service provider can access a Trusted Platform Module (TPM) through an Application Program Interface (API) request. In an unprotected mode, the BAR is programmed so that users can access the configuration information, but the TPM is blocked. Blocking of the TPM is achieved by changing a configuration file, which changes an overall image of the card. With the modified image not matching an original image, the TPM blocks access to data, such as encryption keys. Separate interfaces can be used for user access (PCIe) and service provider access (Ethernet) to the server computer. The server computer can then be toggled back to the protected mode by switching the configuration file to the original configuration file.
-
公开(公告)号:US20210058301A1
公开(公告)日:2021-02-25
申请号:US17091811
申请日:2020-11-06
Applicant: Amazon Technologies, Inc.
Inventor: Anthony Nicholas Liguori , Matthew Shawn Wilson
IPC: H04L12/24 , G06F9/455 , G06F9/4401
Abstract: At a network manager of an extension resource group of a provider network, a message comprising a command to launch a compute instance is received at an address which is part of a first network configured at a premise external to the provider network. The extension resource group includes a first host at the external premise. Within a second network configured at the external premise, the first host is assigned an address within a second address range. Addresses within the second range are also assigned to hosts within the provider network. The command is transmitted to the first host, and a compute instance is instantiated.
-
公开(公告)号:US10911371B1
公开(公告)日:2021-02-02
申请号:US14658769
申请日:2015-03-16
Applicant: Amazon Technologies, Inc.
Inventor: Christopher Richard Jacques De Kadt , Matthew Shawn Wilson , John Merrill Phillips , Michael Groenewald , Jacobus Johannes Van Der Merwe
IPC: G06F15/173 , G06F15/177 , H04L12/24 , H04L29/06 , H04L29/08 , H04L9/00 , H04L12/911 , G06F9/455
Abstract: A control-plane component of a network-accessible service receives an indication that a resource request has been submitted by a client. The component determines, based at least in part on metadata associated with the client, that a first parameter selection policy (PSP) of one or more PSPs created for the service is applicable. The first PSP includes a condition set to determine applicability, and a value set indicating parameter values to be used to fulfill resource requests to which the PSP applies. Using the value set, an effective parameter value list for one or more operations to be performed to allocate the requested resource is generated, and the allocation operations are initiated.
-
公开(公告)号:US10860305B1
公开(公告)日:2020-12-08
申请号:US15721307
申请日:2017-09-29
Applicant: Amazon Technologies, Inc.
Inventor: Jason Alexander Harland , Matthew Shawn Wilson , Darin Lee Frink , Nafea Bshara , Ziv Harel , Anthony Nicholas Liguori
IPC: G06F9/44 , G06F8/65 , G06F13/40 , G06F9/4401 , G06F9/54 , H03K19/177
Abstract: A server includes a motherboard and a programmable logic device coupled to the motherboard. The server also includes a hardware device coupled to the motherboard and the programmable logic device. The server further includes a non-volatile memory storing firmware for the hardware device. The non-volatile memory is coupled to the motherboard and the programmable logic device. The server further includes a peripheral device coupled to the motherboard and the programmable logic device. The peripheral device receives firmware data from a management server. The peripheral device verifies that the firmware data corresponds to the hardware device. The peripheral device further holds the hardware device in reset mode. The peripheral device stores the firmware data on the non-volatile memory to update the firmware and releases the hardware device from reset mode after updating the firmware.
-
公开(公告)号:US10853129B1
公开(公告)日:2020-12-01
申请号:US16358355
申请日:2019-03-19
Applicant: Amazon Technologies, Inc.
Inventor: Sudipta Sengupta , Haifeng He , Pejus Manoj Das , Poorna Chand Srinivas Perumalla , Wei Xiao , Shirley Xue Yi Leung , Vladimir Mitrovic , Yongcong Luo , Jiacheng Guo , Stefano Stefani , Matthew Shawn Wilson
Abstract: Implementations detailed herein include description of a computer-implemented method to migrate a machine learning model from one accelerator portion (such as a portion of a graphical processor unit (GPU)) to a different accelerator portion. In some instances, a state of the first accelerator portion is persisted, the second accelerator portion is configured, the first accelerator portion is then detached from a client application instance, and at least a portion of an inference request is performed using the loaded at least a portion of the machine learning model on the second accelerator portion that had been configured.
-
-
-
-
-
-
-
-
-