Amazon Technologies, Inc.

Antoun Joubran Kanawati

Vishal Parakh

Mukul Vijay Karnik

Ajay Gopalakrishnan

Omer Baluch

Si Yin

Kiran-Kumar Muniswamy-Reddy

Pradeep Vincent

Diwakar Chakravarthy

Melissa Elaine Davis

Jamie Hunter

Rachit Chawla

Alexander Julian Tribble

Julio Delgado Mangas

Timothy Allen Gilman

Aseem Cheema

Parth Patel

Manoj Kumar

Roshith Rajagopal

Kal Lyndon McFate

Gary Paul Raden, Ii

Timothy Andrew Rath

Joshua Abel Shagam

Jared Scott Lundell

Bhupinder Singh Sidana

Kiran-Kumar Muniswamy Reddy

A resource management service implements techniques for provisioning a process with computing resources sufficient to process a query. A query is received and computing system resources sufficient to process the query are provisioned. A response to the query is generated by running the process with the provisioned computing system resources.

Resource management service

Prediction logic analyzes previous data usage activities of a customer process running on a host machine to generate a first prediction indicating that the customer process will request a first data set at a first time. The prediction logic retrieves the first data set from long-term storage and loads the first data set into memory on the host machine in advance of the first time in order to provide the customer process with access to first data set in the memory during a period between the first time and a second time. The prediction logic further generates a second prediction indicating that the customer process will not access the first data set for a threshold period of time after the second time and stores the first data set in the long-term storage at the second time.

Predictive loading and unloading of customer data in memory

At a recovery manager associated with a cluster, a determination is made as to whether a replacement for a first node of the cluster can be elected by the other nodes of the cluster using a first election protocol. The recovery manager selects a second node of the cluster as a replacement for the first node, based on data item replication progress made at the node, and transmits an indication that the second node has been selected to one or more nodes of the cluster.

Hybrid cluster recovery techniques

A data analytics system may receive query definitions from which relationships between datasets may be identified. The query definitions may be analyzed to determine estimated costs and frequencies of combining a first and second dataset. Based on the cost and frequency, a combined dataset may be generated to by joining data from the first and second datasets. The combined dataset may be stored. Queries that comprise instructions to combine the first and second datasets may be processed by instead accessing the combined dataset.

Analysis engine data intake

Processes and systems are disclosed for selecting a producer system from a number of producer systems to lease to a consumer system. A leasing agent, in response to a request from the consumer system for access to a service at a producer system, can identify a producer system to lease to the lease requestor based, at least in part, on a selection weight associated with each producer system that the leasing agent is assigned. The selection weights can be modified based on status information associated with each of the producer systems. This status information may be obtain from the producer systems and/or from a consumer system that has previously accessed the producer system. The consumer system may provide the status information to the leasing agent as part of the consumer system's lease request.

Producer system registration

Embodiments of a distributed caching system are disclosed that cache data across multiple computing devices on a network. In one embodiment, a first caching system serves as a caching front-end to a distributed cluster of additional caching systems. The caching systems may be spread over multiple partition groups. In one embodiment, cache writes at a cache system in one partition group are distributed to other partition groups. By propagating the cache writes across multiple partition groups, the caches at the different partition groups include more recently accessed data, thereby increasing the likelihood of cache hits.

Distributed caching system

Processes are disclosed for decreasing contention in caches in order to increase the efficiency of multi-threaded or multi-processor systems. By using multiple locks in a cache, smaller portions of the cache can be locked during cache updates (e.g., during a data update or a storage block eviction). As only small portions of the cache are locked at any given time, contention between threads, particularly in multi-processor implementations, will likely be reduced. For example, if different threads are trying to update different entries in the cache, the threads can proceed with updating the cache concurrently.

Multi-lock caches

Processes are disclosed for embodiments of a caching system to utilize a snapshot file or other limited size data structure to store a portion of the data stored in a cache. The snapshot file can be stored on persistent or otherwise non-transitory storage so that, even in case of a restart, crash or power loss event, the data stored in the snapshot file persists and can be used by the caching system after starting up. The snapshot file can then be used to restore at least some data into the cache in cases where the cached data in the cache is lost. For example, in cases of a cold-start or restart, the caching system can load data from the snapshot file into the empty cache. This can increase the number of cache hits since the cache is repopulated with useful data at startup.

Cached data replication for cache recovery

Producer system selection

Systems and methods aggregate data and use caching techniques. In one implementation, a computer-implemented method aggregates data. A server receives a request for data that includes one or more identifiers. Based on at least one of the one or more identifiers, at least one template defining a view of the data is retrieved. One or more modules are identified based on the at least one template and one or more services are invoked to obtain the one or more modules. The one or more modules are received and processing by applying one or more business logic rules. At least a portion of the one or more processed modules are stored in a data store. The view of the data is generated based on the processed one or more modules and a response is transmitted to a client that includes the view of the data.

Data aggregation and caching

Embodiments of a distributed caching system are disclosed that cache data across multiple computing devices on a network. In one embodiment, a first cache system serves as a caching front-end to a distributed cluster of additional cache systems. The first cache system can distribute cache requests to the additional cache systems. The first distributed caching system can also serve as a cache server itself, by storing data on its own internal cache. For example, the first cache system can first attempt to find a requested data item on the internal cache, but, if the lookup results in a cache miss, the first cache system can search the additional cache systems for the data. In some embodiments, the first cache system is configured to identify additional requests for the first data item and duplicate the requested data item in order to respond to the additional requests, which allows for greater responsiveness to requests.

Processes and systems are disclosed for a remote messaging protocol that combines application data and reliability information into a three-packet handshake exchange. Each packet may comprise message information indicating an initial packet, or an acknowledgement packet, along with a unique identifier for identifying responses to the initial message. Time-to-live and retransmission timers may be used in order to increase reliability of the protocol.

Growing community of inventors

Seattle, WA, United States of America

Antoun Joubran Kanawati