Thoughtspot, Inc.

Alcatel Lucent

Instart Logic, Inc.

Akamai Technologies, Inc.

Wsou Investments, LLC

Ashok Anand

Mansoor A Alicherry

Bhanu Prakash

Tushar Marda

Kapil Khurana

Shoban Preeth Chandrabose

Manjunath Bharadwaj Subramanya

Hariharan Kolam

Amit Prakash

Mohammad H Reshadi

Prateek Gaur

Sanjay Agrawal

Sharad Jaiswal

Satyam Shekhar

Donko Donjerkovic

Anand Kodaganur

Akshay Agrawal

Kedar Milind Kulkarni

Sujay Alur Narayana

Yellamelli Bharath Mourya

Manish Agarwal

Shrirang Ballal

Ambareesh Sreekumaran Nair Jayakumari

Preet A Shah

Robert Hou

Vibhav Sarraf

Dhwani Kotecha

Krishna P Puttaswamy Naga

Bowei Du

Samprit Biswas

Bishal Singh

Divyam Lamiyan

Shabon Preeth Chandrabose

Aditya Akella

Updating a constituent-data index includes determining whether sampling is currently contraindicated for the column, in response to determining that sampling is currently contraindicated for the column, omitting sampling the column, in response to determining that sampling is currently other than contraindicated for the column, sampling the column, determining whether updating the constituent-data index is currently contraindicated for the column, in response to determining that updating the constituent-data index is currently contraindicated for the column, omitting updating the constituent-data index with respect to the column, and in response to determining that updating is currently other than contraindicated for the column, updating the constituent-data index with respect to the column.

Indexing constituent data

Automatic data modeling in includes identifying an analytical object in response to first data expressing usage intent, generating an analytical model generation data query for the analytical object, obtaining a trained analytical model generated in accordance with the analytical model generation query and trained using results data obtained in accordance with the analytical object, generating a resolved request representing second data expressing usage intent and indicating a request for results data obtained using the trained analytical model, generating an analytical model results data query for obtaining the results data in accordance with the trained analytical model and the analytical object, and outputting data for presenting a visualization of the results data obtained by executing the analytical model results data query, wherein a first portion of the results data corresponds with the analytical object and a second portion of the results data corresponds with the trained analytical model.

Automatic machine learning data modeling in a low-latency data access and analysis system

A request for database results is received from a query coordinator at a database instance of a distributed database. The request includes a query execution instruction of a query plan and an indication of override instructions corresponding to the query execution instruction. The override instructions are such that they do not modify the query plan. The database instance includes the override instructions in a set of high-level language query instructions. The database instance performs just-in-time compiling of the set of high-level language query instructions to obtain a machine language query for performing the query execution instruction of the query plan. The database instance executes the machine language query to obtain the database results. The database instance then transmits the database results to the query coordinator.

Just-in-time injection in a distributed database

A request to execute a database command is transmitted from a device to a database command execution device. The request includes a first indicator of a first set of tokens available at the device at a time that the request is transmitted. A response to the request is received. The response includes a second indicator of a second set of tokens available at the database command execution device at a time that the request is received at the database command execution device. Responsive to a determination that the second indicator is different from the first indicator, a request for updated tokens is transmitted to the database command execution device. The first set of tokens is updated based on a received response to the request for the updated tokens. A list of tokens the updated first set of tokens that match a partial token received as an input is output.

Secure and efficient database command execution support

Query planning in a distributed database that includes a table partitioned into shards according to a sharding criterion and distributed to database instances includes receiving a data-query. The data-query includes a 'distinct count' clause on a first column and a “group by” clause on least a second column. A query plan is formulated to include respective instructions for converting, at at least some of the database instances, distinct values of the first column grouped by values of the second column into a count of the distinct values grouped by the values of the second column to obtain respective intermediate results; instructions for receiving the respective intermediate results from at least a subset of the at least some of the database instances; and instructions for concatenating the respective intermediate results using a summing operation to obtain the first “distinct count” of the first column grouped by the second column.

Aggregation operations in a distributed database

Automatic data modeling in a low-latency data access and analysis system includes identifying an analytical-object in response to first data expressing usage intent, generating an analytical model generation data-query for the analytical-object, obtaining a trained analytical model generated in accordance with the analytical model generation query and trained using results data obtained in accordance with the analytical-object, generating a resolved-request representing second data expressing usage intent and indicating a request for results data obtained using the trained analytical model, generating an analytical model results data-query for obtaining the results data in accordance with the trained analytical model and the analytical-object, and outputting data for presenting a visualization of the results data obtained by executing the analytical model results data-query, wherein a first portion of the results data corresponds with the analytical-object and a second portion of the results data corresponds with the trained analytical model.

Injecting override instructions associated with query executions instructions performed on a distributed database includes receiving a data-query; generating, by a first database instance, a query plan that includes a first query execution instruction for transmission to a second database instance; transmitting, by the first database instance, a request for partial results to the second database instance, where the request includes the first query execution instruction and an indication of override instructions corresponding to the first query execution instruction; responsive to a determination that the request includes the indication, including, by the second database instance, the override instructions in a set of high-level language query instructions; obtaining, by the second database instance, a machine language query based on the set; executing, at the second database instance, the machine language query to obtain the partial results; and transmitting, by the second database instance, the partial results to the first database instance.

Obtaining an approximate unique count for a column from a table from a database includes, generating, for a value from an unevaluated row, a hash value in a defined range of hash values, determining a cardinality of leading zeros in the hash value, identifying a bucket with respect to the hash value from a plurality of buckets corresponding to the defined range of hash values, wherein the buckets from the plurality of buckets correspond with respective non-overlapping portions of the defined range of hash values, such that the hash value is in the portion of the defined range of hash values corresponding to the bucket, and appending to an unsorted sparse representation a bucket identifier for the bucket and the cardinality of the leading zeros, and, in response to a determination that unevaluated rows are unavailable in the table, determining the approximate unique count using the unsorted sparse representation.

Approximate unique count

Querying a distributed database including a table sharded into shards distributed to database instances includes receiving a data-query that includes an aggregation clause on a first column and a grouping clause on a second column; obtaining and outputting results data. Obtaining the results data includes receiving, by a query coordinator, intermediate results data; and combining, by the query coordinator, the intermediate results to obtain the results data. Receiving the intermediate results data includes receiving, from a first database instance, first aggregation values indicating, on a per-group basis in accordance with the grouping clause, a respective aggregation value of distinct values of the first column in accordance with the aggregation clause, and receiving, from a second database instance, second aggregation values indicating, on a per-group basis in accordance with the grouping clause, a respective aggregation value of distinct values of the first column in accordance with the aggregation clause.

Enabling auto-completion of database commands includes receiving, at a database command execution device from a client device, a request to execute a database command where the request includes a first indicator of a first set of tokens of the database that is available at the client device; executing the database command; transmitting, from the database command execution device to the client device, a response to the request that includes a status of the execution of the database command and a second indicator of a second set of tokens of the database that is different from the first set of tokens; receiving, at the database command execution device and from the client device, an update-tokens request that includes the first indicator; and transmitting, from the database command execution device and to the client device, data indicative of differences between the second set of tokens and the first set of tokens.

Systems and methods for multi-layered key-value storage are described. For example, methods may include receiving two or more put requests that each include a respective primary key and a corresponding respective value; storing the two or more put requests in a buffer in a first datastore; determining whether the buffer is storing put requests that collectively exceed a threshold; responsive to the determination that the threshold has been exceeded, transmitting a write request to a second datastore, including a subsidiary key and a corresponding data file that includes the respective values of the two or more put requests at respective offsets in the data file; for the two or more put requests, storing respective entries in an index in the first datastore that associate the respective primary keys with the subsidiary key and the respective offsets; and deleting the two or more put requests from the buffer.

Multi-layered key-value storage

Described are methods and systems for improved cardinality estimation. A method may include obtaining a data-query, obtaining a row, generating a hash value, determining a cardinality of leading zeros in the hash value, identifying a bucket with respect to the hash value, including a bucket identifier and the cardinality of leading zeros in a representation, determining the approximate unique count, and outputting the approximate unique count as results data responsive to the portion of the data-query.

Data-query execution with distributed machine-language query management in a low-latency database analysis system may include obtaining, at a distributed in-memory database, a data-query expressing a request for data in a defined structured query language associated with the distributed in-memory database, automatically generating a high-level language query representing at least a portion of the data-query, obtaining a machine language query corresponding to the high-level language query, executing the machine language query to obtain results data, and outputting the results data. Obtaining the machine language query may include determining whether the machine language query is cached, and in response to a determination that the machine language query is unavailable, sending a request for the machine language query to a distributed machine-language-query management instance.

Machine language query management for low-latency database analysis system

A method of speeding up the delivery of a dynamic webpage is disclosed. A plurality of responses to a plurality of requests for a webpage is received, the webpage including dynamic components. The plurality of responses is compared to identify common and dynamic components across the plurality of requests. A cached stub is dynamically adapted based on the comparison within a learning window, the cached stub including the identified common components and placeholders for portions of the identified dynamic components. The dynamically adapted cached stub is sent in response to at least some of the plurality of requests for the webpage. A frequency of reloading of the webpage is monitored, wherein a reloading of the webpage is triggered by a detection of the dynamically adapted cached stub having one of a plurality of types of error. The learning window is adjusted based on a frequency of reloading of the webpage.

Growing community of inventors

Bengaluru, India

Ashok Anand