公开(公告)号：US20190196748A1

公开(公告)日：2019-06-27

申请号：US15850578

申请日：2017-12-21

Applicant: Microsoft Technology Licensing, LLC

Inventor： Anirudh BADAM ,  Badriddine KHESSIB ,  Laura Marie CAULFIELD ,  Mihail Gavril TARTA ,  Robin Andrew ALEXANDER ,  Xiaozhong XING ,  Zhe TAN ,  Jian XU

IPC: G06F3/06 ,  G06F9/54 ,  G06F9/455

CPC classification number: G06F3/0667 ,  G06F3/0605 ,  G06F3/0613 ,  G06F3/0631 ,  G06F3/064 ,  G06F3/0688 ,  G06F9/455 ,  G06F9/545 ,  G06F12/0246 ,  G06F2009/45587 ,  G06F2009/45595 ,  G06F2212/7201

Abstract: A system includes reception of a request from a first application to create a virtual open-channel solid state drive associated with a first bandwidth and first capacity, association, in response to the request, of block addresses of a virtual address space of the first application with block addresses of one or more blocks of a first one of a first plurality of channels of a first open-channel solid state drive and with block addresses of one or more blocks of a second one of a second plurality of channels of a second open-channel solid state drive, reception, from the first application, of a first I/O call associated with one or more block addresses of the virtual address space, determination of block addresses of one or more blocks of the first one of the first plurality of channels which are associated with the one or more block addresses of the virtual address space, and execution of the first I/O call on the determined block addresses of one or more blocks of the first one of the first plurality of channels.

公开(公告)号：US20250060998A1

公开(公告)日：2025-02-20

申请号：US18452326

申请日：2023-08-18

Applicant: Microsoft Technology Licensing, LLC

Inventor： Amar PHANISHAYEE ,  . Ankit ,  Deepak NARAYANAN ,  Mihail Gavril TARTA

IPC: G06F9/50

Abstract: Systems and methods for optimizing thread allocation in a model serving system include estimating a batch size for inference requests. An optimal configuration is then determined that defines a number of inference instances, a number of threads per inference instance, and a sub-batch size per inference instance for processing a batch of inference requests of the batch size using intra-operator parallelism that minimizes average per-batch latency. The optimal configuration is determined with reference to a plurality of predetermined model profiles that define single-inference average batch latencies for different combinations of thread counts and batch sizes, the predetermined model profiles being used as input to a dynamic programming algorithm that identifies optimal configurations that minimize the average per-batch latency.

公开(公告)号：US20190108166A1

公开(公告)日：2019-04-11

申请号：US16209771

申请日：2018-12-04

Applicant: MICROSOFT TECHNOLOGY LICENSING, LLC

Inventor： Mihail Gavril TARTA ,  Gopal KAKIVAYA ,  Preetha Lakshmi SUBBARAYALU

IPC: G06F16/23 ,  G06F16/27

Abstract: In various embodiments, methods and systems for optimizing database transactions based on replicable differential data store data structure are provided. A write operation request, having a key for a write operation on a replicable differential store data structure, is accessed. An intent write lock on a differential state and a write lock on the key are acquired. The differential state comprises a result set of currently committing transactions. A transaction instance, of the write operation, is generated for a write set, the transaction instance comprising a modification to the key. The write-set comprises an uncommitted set of writes for in-flight transactions. A determination is made that the write operation is committed. A result of the transaction instance is persisted when the write operation is committed. It is contemplated that the differential state and a consolidated state can be merged, the consolidated state comprises a result set of previously committed transactions.

公开(公告)号：US20190303047A1

公开(公告)日：2019-10-03

申请号：US16447094

申请日：2019-06-20

Applicant: Microsoft Technology Licensing, LLC

Inventor： Anirudh BADAM ,  Badriddine KHESSIB ,  Laura Marie CAULFIELD ,  Mihail Gavril TARTA ,  Robin Andrew ALEXANDER ,  Xiaozhong XING ,  Zhe TAN ,  Jian XU

IPC: G06F3/06 ,  G06F9/54 ,  G06F9/455 ,  G06F12/02

Abstract: A system includes reception of a request from a first application to create a virtual open-channel solid state drive associated with a first bandwidth and first capacity, association, in response to the request, of block addresses of a virtual address space of the first application with block addresses of one or more blocks of a first one of a first plurality of channels of a first open-channel solid state drive and with block addresses of one or more blocks of a second one of the first plurality of channels, reception, from the first application, of a first I/O call associated with one or more block addresses of the virtual address space, determination of block addresses of one or more blocks of the first one of the first plurality of channels which are associated with the one or more block addresses of the virtual address space, and execution of the first I/O call on the determined block addresses of one or more blocks of the first one of the first plurality of channels.

公开(公告)号：US20170052856A1

公开(公告)日：2017-02-23

申请号：US14829044

申请日：2015-08-18

Applicant: Microsoft Technology Licensing, LLC

Inventor： Preetha Lakshmi SUBBARAYALU ,  Gopala Krishna R. KAKIVAYA ,  Mihail Gavril TARTA

IPC: G06F11/20

CPC classification number: G06F11/2023 ,  G06F11/00 ,  G06F2201/805

Abstract: A state manager provides transactional distributed lifecycle management of a group of different application-level state providers, namely, differently structured application program data structures. The state providers are atomic with respect to one another. The state provider is replicated to one or more secondary nodes of a distributed network. The state providers are persistent despite one or more node operational failures. State provider lifecycle operations include creation of a transactional distributed state provider as a member of a group of different application-level state providers which include differently structured application program data structures, deletion of a previously created transactional distributed state provider, and/or enumeration of any previously created transactional distributed state providers. A given state provider may be read or written by one or more applications. Implementation restrictions and other avoidance conditions are satisfied in particular cases.

Abstract translation: 状态管理器提供一组不同应用级状态提供者的事务性分布式生命周期管理，即不同结构化的应用程序数据结构。 国家提供者是彼此原子的。 状态提供者被复制到分布式网络的一个或多个次要节点。 尽管有一个或多个节点操作失败，状态提供程序仍然持续存在。 状态提供商生命周期操作包括创建事务性分布式状态提供者作为不同应用级状态提供者的成员，其包括不同结构化的应用程序数据结构，删除先前创建的事务性分布式状态提供者，和/或枚举任何 先前创建的事务分布式状态提供程序。 给定的状态提供者可以被一个或多个应用程序读取或写入。 在特定情况下，满足实施限制和其他回避条件。