Microsoft Technology Licensing, LLC

Arnd Christian König

Vivek Ravindranath Narasayya

Ajay Kalhan

John Morgan Oslake

Shize Xu

Junfeng Dong

Jiaqi Liu

Changsong Li

Manoj A Syamala

Steven Mark Drucker

Mark Steven Manasse

Danyel Aharon Fisher

Sudipto Das

Feng Li

Eric David Brill

Cesar Alejandro Galindo-Legaria

Paul Adrian Oltean

Kenneth Ward Church

Martin Miroslavov Markov

Yi Shan

Aarati Kakaraparthy

Tobias Ziegler

Manoj Syamala

Mahadevan Sankara Subramanian

Chi Yang

Ye Eun Park

Dang Hai Pham

Kabita Mahapatra

Sumeet Priyadarshee Dash

Intelligent memory brokering for multiple process instances, such as relational databases (e.g., SQL servers), reclaims memory based on value, thereby minimizing cost across instances. An exemplary solution includes: based at least on a trigger event, determining a memory profile for each of a plurality of process instances at a computing node; determining an aggregate memory profile, the aggregate memory profile indicating a memory unit cost for each of a plurality of memory units; determining a count of memory units to be reclaimed; identifying, based at least on the aggregate memory profile and the count of memory units to be reclaimed, a count of memory units to be reclaimed within each process instance so that a total cost is minimized to reclaim the determined count; and communicating, to each process instance having identified memory units to be reclaimed, a count of memory units to be reclaimed within the process instance.

Memory management for multiple process instances

Placement of a tenant database in an oversubscribed, database-as-a-service cluster comprised of a plurality of nodes is described. The placement may be based on per-node estimates of a probability of resource demand violation if the tenant database is placed on the node. Past resource usage of similar tenant databases subscribed to the cluster that are collected and stored as compressed traces may be used to obtain the estimates. In some examples, based on the estimates, a per-node expected number of resource violations is determined and compared across nodes, where the determined placement minimizes the number of resource violations. In other examples, when the tenant database is being placed in parallel with other tenant databases, a score assigned to each valid configuration for the placement may be modified based on the estimates, where the determined placement is the configuration having a lowest score.

Tenant database placement in oversubscribed database-as-a-service cluster

Methods for distributed histogram computation in a framework utilizing data stream sketches and samples are performed by systems and devices. Distributions of large data sets are scanned once and processed by a computing pool, without sorting, to generate local sketches and value samples of each distribution. The local sketches and samples are utilized to construct local histograms on which cardinality estimates are obtained for query plan generation of distributed queries against distributions. Local statistics of distributions are also merged and consolidated to construct a global histogram representative of the entire data set. The global histogram is utilized to determine a cardinality estimation for query plan generation of incoming queries against the entire data set. The addition of new data to a data set or distribution involves a scan of the new data from which new statistics are generated and then merged with existing statistics for a new global histogram.

Distributed histogram computation framework using data stream sketches and samples

Methods for determining cache activity and for optimizing cache reclamation are performed by systems and devices. A cache entry access is determined at an access time, and a data object of the cache entry for a current time window is identified that includes a time stamp for a previous access and a counter index. A conditional counter operation is then performed on the counter associated with the index to increment the counter when the time stamp is outside the time window or to maintain the counter when the time stamp is within the time window. A counter index that identifies another counter for a previous time window where the other counter value was incremented for the previous cache entry access causes the other counter to be decremented. A cache configuration command to reclaim, or additionally allocate space to, the cache is generated based on the values of the counters.

System and method for determining cache activity and optimizing cache reclamation

A solution is disclosed for memory management of serverless databases that includes: based at least on detecting a trigger event, determining whether memory is to be reclaimed; based at least on determining that memory is to be reclaimed, determining an amount of memory to be reclaimed; identifying memory to be reclaimed; and reclaiming the identified memory. Disclosed solutions are flexible, enabling customization of the aggressiveness and manner of memory reclamation. This permits users to specify a tailored balance point between performance and cost, for arrangements that bill users based on resource usage (e.g., memory consumed by a serverless database). In some examples, users specify a ramp-down parameter that is used to determine how much memory can be evicted in a particular reclamation event, time intervals (or another criteria) for triggering a reclamation event, and a definition for whether a cache is active.

Memory management for serverless databases

Architecture that enables a Database-as-a-Service (DaaS) to auto-scale container sizes on behalf of tenants. An abstraction is provided that enables tenants to reason about monetary budget and query latency, rather than resource provisioning. An auto-scaling module automatically determines a container size for a subsequent billing interval based on telemetry that comprises latencies (e.g., waits), resource utilizations, and available budget, for example. A set of robust signals are derived from database engine telemetry and combined to significantly improve accuracy of resource demand estimation for database workloads. In a more specific implementation, resource demands can be estimated for arbitrary SQL (structured query language) workloads in a relational database management system (RDBMS).

Automatic demand-driven resource scaling for relational database-as-a-service

Incremental query results and confidence interval values associated with respective incremental query results may be obtained. Visualization shape objects indicating uncertainty values may be determined, based on mapping values of respective incremental query results and confidence interval values to points in the associated visualization shape objects, the uncertainty values visualized based on proportional shapes of the visualization shape objects. At least one visualization comparison object representing a comparison of a plurality of distributions associated with the obtained incremental query results and confidence interval values may be determined. Display of the plurality of visualization shape objects and the at least one visualization comparison object may be initiated.

Visualization of changing confidence intervals

Hash values corresponding to a file are processed in windows to determine a minimum hash value for each window. Each window may begin at a minimum hash value determined for a previous window and end after a fixed number of hash values. If a hash value is less than a threshold hash value, it is added to a buffer that is used to store the hash values in sorted order for a current window. If a hash value is greater than the threshold, it is added to another buffer whose hash values are not stored in sorted order. At the end of the current window, the minimum hash value in the first buffer is selected as the landmark for the window. If the first buffer is empty, then the hash values in the other buffer are sorted and the minimum hash value is selected as the landmark for the window.

Determination of landmarks

The subject disclosure is directed towards simulating query execution to provide incremental visualization for a global data set. A data store may be configured for searching at least a portion of a global data set being stored at an enterprise-level data store. In response to a user-issued query, partial query results are provided to a front-end interface for display to the user. The front-end interface also provides statistical information corresponding to the partial query results in relation to the global data set, which may be used to determine when a current set of query results becomes acceptable as a true/accurate estimate.

Incremental visualization for structured data in an enterprise-level data store

A system that facilitates selecting advertisements that match a search query is described herein. The system includes a search query receiver component that receives a search query including keywords. The system also includes a match component that uses an associative data structure to identify in the associative data structure one or more data nodes that are associated in the associative data structure with respective unique keys corresponding to respective one or more hashes of combinations of the keywords in the search query. For each identified data node, the match component selects advertisements associated with bid phrases stored in the identified data node that respectively only include keywords included in the search query.

Estimating document similarity using bit-strings

A unique multi-stage classification system and method that facilitates reducing human resources or costs associated with text classification while still obtaining a desired level of accuracy is provided. The multi-stage classification system and method involve a pattern-based classifier and a machine learning classifier. The pattern-based classifier is trained on discriminative patterns as identified by humans rather than machines which allow a smaller training set to be employed. Given humans' superior abilities to reason over text, discriminative patterns can be more accurately and more readily identified by them. Unlabeled items can be initially processed by the pattern-based classifier and if no pattern match exists, then the unlabeled data can be processed by the machine learning classifier. By employing the classifiers in this manner, less human involvement is required in the classification process. Even more, classification accuracy is maintained and/or improved.

Growing community of inventors

Kirkland, WA, United States of America

Arnd Christian König