Abstract: Novel techniques are described for negative entity-aware augmentation using a two-stage augmentation to improve the stability of the model to entity value changes for intent prediction. In some embodiments, a method comprises accessing a first set of training data for an intent prediction model, the first set of training data comprising utterances and intent labels; applying one or more negative entity-aware data augmentation techniques to the first set of training data, depending on the tuning requirements for hyper-parameters, to result in a second set of training data, where the one or more negative entity-aware data augmentation techniques comprise Keyword Augmentation Technique (“KAT”) plus entity without context technique and KAT plus entity in random context as OOD technique; combining the first set of training data and the second set of training data to generate expanded training data; and training the intent prediction model using the expanded training data.

Abstract: Novel techniques are described for positive entity-aware augmentation using a two-stage augmentation to improve the stability of the model to entity value changes for intent prediction. In one particular aspect, a method is provided that includes accessing a first set of training data for an intent prediction model, the first set of training data comprising utterances and intent labels; applying one or more positive data augmentation techniques to the first set of training data, depending on the tuning requirements for hyper-parameters, to result in a second set of training data, where the positive data augmentation techniques comprise Entity-Aware (“EA”) technique and a two-stage augmentation technique; combining the first set of training data and the second set of training data to generate expanded training data; and training the intent prediction model using the expanded training data.

Abstract: Disclosed systems relate to propagating tracing across a distributed software application. In an example, an application logs tracing data for a web page from an original server. The application determines, at the web browser, that the web page from the original server requires a request to an external server outside a domain of the original server. The application interrogates the external server to determine whether the external server allows tracing headers in requests. The application updates an allow list, at the web browser, to indicate that the external server allows tracing headers in requests from the original server. The application injects a tracing header in the request based on a result of the interrogation. The application sends, from the web browser to the external server, the request with the tracing header. The external server is configured to log tracing data based on the tracing header.

Abstract: Novel techniques are described for data augmentation using a two-stage entity-aware augmentation to improve model robustness to entity value changes for intent prediction.

Abstract: Techniques for persisting user data across secure shell instances are provided. A method includes receiving a first request from a session manager service to establish a connection to a secure shell instance and restore a user block volume with corresponding backup user data. The method may include reserving an empty block volume. The method may also include transmitting a backup data identifier associated with the corresponding backup user data to a backup service and receiving the corresponding backup user data from the backup service. The method may further include providing the corresponding backup user data to the empty block volume to create a restore volume and transmitting a restore volume identifier corresponding to a data center identifier to the session manager service. The method may include receiving a second request to attach the restore volume to a reserved instance, the second request being received from the session manager service.

Abstract: Machine learning techniques are disclosed for predicting a task event such as a service completion event based on a predefined workflow. In one aspect a method includes obtaining initial data for a service request (e.g., an account application), enriching the initial data with data from one or more repositories of an enterprise executing the service request, generating a data structure comprising independent variables extracted from the enriched data, receiving a request for a prediction of a completion time for the service request (e.g., an account opening event) at a first time during processing of the service request in accordance with each workflow, in response to receiving the request for the prediction, inputting the data structure into a machine-learning regression model, predicting, using the machine-learning regression model, a completion time for the service request, and providing the completion time for the service request.

Abstract: Cyber-security techniques are described for monitoring a cloud environment and identifying potential problems, including malicious threats, to the monitored cloud environment using operational telemetry. Techniques are described for monitoring and collecting data related to reverse or recursive DNS (rDNS) traffic associated with a monitored cloud environment. The recursive DNS traffic includes recursive DNS (rDNS) requests originating from the cloud environment and responses to those requests received from DNS resolvers. This collected data is then analyzed to identify potential threats to the monitored cloud environment. The collected data may be analyzed to identify potential sources of threats and to identify one or more portions of the cloud environment that are the targets of the threats. The analysis may trigger alerts to be generated, actions to be performed (e.g., protective measures), reports to be generated, patterns to be recognized, etc.

Abstract: Techniques are described for performing different types of restart operations for a file storage replication between a source file system and a target file system in different cloud infrastructure regions. In certain embodiments, the disclosed techniques perform a restart operation to terminate a current cross-region replication by synchronizing resource cleanup operations in the source file system and the target file system, respectively. In other embodiments, disclosed techniques perform a restart operation to allow a customer to reuse the source file system by identifying a restartable base snapshot in the source file system without dependency on the target file system.

Abstract: Novel techniques are described for performing hierarchical key management involving an end-to-end file storage replication between different cloud infrastructure regions. The novel hierarchical key management comprises three different keys, a first security key for the source region, a session key, valid only for a session, for the transfer of data between two different regions, and a second security key for the target region. Novel techniques are also described for using different file keys for different files of a file system in each region.

Abstract: Plugins that are independently written are executed in a collaborative manner to analysis a log. A plugin executing with respect to a particular node of a hierarchical data structure determines values for a set of keys based on information of the particular node and/or any ancestor nodes, and information stored in a shared repository. The plugin stores the values for the keys as additional information of the particular hierarchical node and/or into the shared repository. The plugin does not access information of non-ancestor nodes when executing with respect to the particular hierarchical node. Each plugin writes into and retrieves from the shared repository using the shared naming convention, thereby sharing information. The sequence of execution of the plugins is not dependent on dependencies amongst the plugins. If a dependent plugin requiring an output from a requisite plugin is first executed, the dependent plugin is flagged as pending and subsequently re-executed.

Abstract: Techniques are described for ensuring end-to-end fair-share resource utilization during cross-region replication. In certain embodiments, a fair-share architecture is used for communication among pipeline stages performing a cross-region replication between different cloud infrastructure regions. Cross-region replication-related jobs are distributed evenly from a pipeline stage into a temporary buffer in the fair-share architecture, and then further distributed evenly form the fair-share architecture to parallel running threads of next pipeline stage for execute. Techniques for static and dynamic resource allocations are also disclosed.

Abstract: A method, apparatus, and product to provide a parser for property graph queries with precise error reporting and auto-completion based on information from property graph schemas. The approach generally comprises analysis of graph queries prior to their execution to identify issues prior to execution. In some embodiments, the approach includes any of: use of a property graph schema to determine whether names in a received property graph query exist within a corresponding property graph; determining whether the property graph query includes a comparison of mismatched data types; providing an autocomplete suggestion feature for assistance in resolving errors or corresponding to a cursor position within a query string; or evaluation of a property graph query to determine whether it would return an empty result. In some embodiments, property graph query analysis is performed using a context aware approach.

Abstract: Plugins that are independently written are executed in a collaborative manner to analysis a log. A plugin executing with respect to a particular node of a hierarchical data structure determines values for a set of keys based on information of the particular node and/or any ancestor nodes, and information stored in a shared repository. The plugin stores the values for the keys as additional information of the particular hierarchical node and/or into the shared repository. The plugin does not access information of non-ancestor nodes when executing with respect to the particular hierarchical node. Each plugin writes into and retrieves from the shared repository using the shared naming convention, thereby sharing information. The sequence of execution of the plugins is not dependent on dependencies amongst the plugins. If a dependent plugin requiring an output from a requisite plugin is first executed, the dependent plugin is flagged as pending and subsequently re-executed.

Abstract: The disclosed systems, methods and computer readable media relate to managing Non-Volatile Memory Express (NVMe) over Transmission Control Protocol (TCP) (NVMeOTCP) connections between a smart network interface card (smartNIC) and a block storage data plane (BSDP) of a cloud computing environment. A software agent (“agent”) executing at the smartNIC may manage a number of network paths (active and, in some cases, passive network paths). The agent may monitor the network traffic (e.g., input/output operations (IOPS)) through the paths (e.g., using established NVMeOTCP connections corresponding to the paths). If a condition is met relating to a performance threshold associated with the monitored paths, the agent may increase or decrease the number established NVMeOTCP connections to match real time network conditions.

Abstract: Techniques are described for partitioning B-tree keys of file systems into key ranges for parallel processing in delta generation during file storage replications between file systems in different cloud infrastructure regions. In certain embodiments, a delta generation processing for cross-region replication may utilize a key-range splitting mechanism involving a recursive algorithm that partitions B-tree keys of a source file system into roughly equal-size key ranges. All the partitioned key ranges may be processed in parallel and concurrently by different processing threads, one thread per key range, to improve the performance of the delta generation and achieve scalability.

Abstract: Techniques are described for efficient replication and maintaining snapshot data consistency during file storage replication between file systems in different cloud infrastructure regions. In certain embodiments, snapshot creation and deletion requests that occur during cross-region replications may be temporarily withheld until appropriate times to execute such requests safely, depending on the timing relationship between such requests and cross-region replication cycles.

Abstract: Techniques are described for efficient replication and maintaining snapshot data consistency during file storage replication between file systems in different cloud infrastructure regions. In certain embodiments, provenance IDs are used to efficiently identify a starting point (e.g., a base snapshot) for a cross-region replication process, conserve cloud resources while reducing network and IO traffic.

Abstract: Novel techniques for end-to-end file storage replication and security between file systems in different cloud infrastructure regions are disclosed herein. In one embodiment, a file storage service generates deltas between snapshots in a source file system, and transfers the deltas and associated data through a high-throughput object storage to recreate a new snapshot in a target file system located in a different region during disaster recovery. The file storage service utilizes novel techniques to achieve scalable, reliable, and restartable end-to-end replication. Novel techniques are also described to ensure a secure transfer of information and consistency during the end-to-end replication.

Abstract: Techniques are described for checkpointing multiple key ranges in parallel and concurrently during file storage replications between file systems in different cloud infrastructure regions. In certain embodiments, multiple range threads processing multiple key ranges, one thread per key range, create checkpoints for their respective key ranges in parallel and concurrently after processing a per-determined number of B-tree keys. In certain embodiments, each thread requests a lock from a central checkpoint record and takes turns for updating a status byte while continuing processing the B-tree keys in its responsible key range. In certain embodiments, upon encountering a failure event, either a system crash or a thread failure, each thread restarts its B-tree key processing from a B-tree key after the most recent checkpoint.

Abstract: Techniques for detecting anomalies in live marketing campaign data are disclosed, including: obtaining baseline data associated with one or more digital marketing campaigns; configuring an anomaly detection model to detect anomalies in digital marketing data, based at least on the baseline data; receiving a live stream of a set of digital marketing data associated with a particular digital marketing campaign that is currently being executed; while the particular digital marketing campaign is being executed: applying the anomaly detection model to the set of digital marketing data, to determine if the set of digital marketing data includes an anomaly relative to the baseline data; prior to completion of the particular digital marketing campaign and responsive to determining that the set of digital marketing data includes the anomaly relative to the baseline data, executing an action to address the anomaly.

Abstract: Novel techniques are described for maintaining file consistency between different cloud infrastructure regions during cross-region file storage replication. The novel techniques comprise upon detecting a file deletion status (i.e., a file directory entry version marked as deletion) in a source file system during a replication cycle, retrieving file name from a valid file directory entry of a prior snapshot to construct a temporary file directory entry containing necessary information for a target file system to reconstruct a duplicate file directory entry version marked as deletion. The temporary file directory entry is transferred from the source file system to the target file system during the same replication cycle.

Abstract: Techniques are described for maintaining data consistency when failure events occur during file storage replications between file systems in different cloud infrastructure regions. In certain embodiments, two generation numbers (or different identifications) are assigned to two groups of processed B-tree key-value pairs, one before and one after a failure event, within a key range. In some embodiments, the two generation numbers are assigned to a group of B-tree key-value pairs processed by a failed thread and another group of B-tree key-value pairs processed by a substitute thread taking over the failed thread to avoid potential data corruption.

Abstract: Systems and methods for geometric based flow programming are disclosed herein. The method can include receiving at least one compiled rule at a first Network Virtualization Device (“NVD”), each of the at least one compiled rules can be applicable to a class of packets received by the first NVD for delivery to a Virtualized Network Interface Card (“VNIC”). The method can include receiving a first packet at the first NVD for delivery to a first VNIC, determining with the first NVD that a first rule of the at least one compiled rule is applicable to the first packet, and processing with the first NVD the first packet according to the first rule.

Abstract: Techniques are described for performing different types of restart operations for a file storage replication between a source file system and a target file system in different cloud infrastructure regions. In certain embodiments, the disclosed techniques perform a restart operation to terminate a current cross-region replication by synchronizing resource cleanup operations in the source file system and the target file system, respectively. In other embodiments, disclosed techniques perform a restart operation to allow a customer to reuse the source file system by identifying a restartable base snapshot in the source file system without dependency on the target file system.

Abstract: Techniques are disclosed for generating defeating textual replies. A computer-implemented method generates a first communicative discourse tree from an utterance and as second communicative discourse tree from a candidate answer derived from a search of keywords in the utterance. The method translates the first communicative discourse tree into a first logical formula from and the second communicative discourse tree into a second logical formula. The method determines that the first logical formula is a negation of the second logical formula. The method further provides the candidate answer to a user device responsive to the determining.

Abstract: Techniques are disclosed for augmenting data sets used for training machine learning models and for generating predictions by trained machine learning models. These techniques may increase a number and diversity of examples within an initial training dataset of sentences by extracting a subset of words from the existing training dataset of sentences. The techniques may conserve scarce sample data in few-shot situations by training a data generation model using general data obtained from a general data source.

Abstract: Techniques are disclosed for generating a message stream configured to indicate a source of the various messages within the message stream. In particular, the indicators identify which interface the messages were received at a terminal (e.g., of a mobile handset). The terminal receives messages on various interfaces, and separates the messages received via different interfaces with interface switch indicators within the message stream. In one embodiment, the smart card receives a message stream that includes sets of messages and interface switch indicators therein. The smart card delivers messages from the message stream to a single logical partition of the smart card up until an interface switch indicator is identified in the message stream. From that point, the smart card delivers messages from the message stream to a different logical partition of the smart card up until another interface switch indicator is identified in the message stream.

Abstract: A novel system is described for performing hierarchical named entity recognition (“HNER”) processing that includes identifying categories at different hierarchical levels for a named entity. The HNER system uses a novel architecture comprising an encoder model and a system of trained machine learning (ML) models to perform the HNER processing, where each trained model in the system of ML models corresponds to a particular hierarchical level, and each model is trained to extract one or more named entities and predict a category for each extracted named entity for the corresponding hierarchical level. Novel techniques are also described for training the various models in HNER system including an encoder model and models in the system of models.

Abstract: The present embodiments relate to a CI replication service that can replicate domain data from IDCS control plane to data plane and to all subscribed regions of a domain. For instance, the CI replication service can provide replication of required resources of a domain for AuthN and AuthZ from an IDCS local region to other regions for high availability (e.g., to improve latency). The CI replication service can replicate the resources from a domain's home region to all subscribed regions for local availability of data for workloads running in those regions. Further, when a new region is subscribed for a domain, then the service can bootstrap that domain's data from home region before enabling that region for the domain.

Abstract: NUMA-aware reader-writer locks may leverage lock cohorting techniques that introduce a synthetic level into the lock hierarchy (e.g., one whose nodes do not correspond to the system topology). The synthetic level may include a global reader lock and a global writer lock. A writer thread may acquire a node-level writer lock, then the global writer lock, and then the top-level lock, after which it may access a critical section protected by the lock. The writer may release the lock (if an upper bound on consecutive writers has been met), or may pass the lock to another writer (on the same node or a different node, according to a fairness policy). A reader may acquire the global reader lock (whether or not node-level reader locks are present), and then the top-level lock. However, readers may only hold these locks long enough to increment reader counts associated with them.

Abstract: A universal code editor (UCE) is described that can be used for authoring and editing artifacts associated with multiple different cloud services. A single instance of the UCE and the associated user interface (UI) can be used to select and edit one or more artifacts from different cloud services. The cloud services may be provided by the same cloud service provider or by different cloud service providers. In certain embodiments, the UCE is implemented as part of a cloud-based IDE.

Abstract: Techniques are disclosed for augmenting data sets used for training machine learning models and for generating predictions by trained machine learning models. The techniques generate synthesized data from sample data and train a machine learning model using the synthesized data to augment a sample data set. Embodiments selectively partition the sample data set and synthesized data into a training data and a validation data, which are used to generate and select machine learning models.

Abstract: The present disclosure describes an anomaly detection system that generates a resource group including a plurality of resources of a monitored environment based on a grouping property. The values of the grouping property associated with the plurality of resources satisfy a first condition. A first invariance identifying property is selected from a set of invariance identifying properties. It is determined whether values of the first invariance identifying property associated with the plurality of resources satisfy a second condition. Responsive to a successful determination, a first invariant is incorporated in a baseline, wherein the first invariant is defined by the grouping property and the first invariance identifying property. The baseline is used by the anomaly detection system for performing anomaly detection of the monitored environment.

Abstract: Techniques are disclosed for providing high performant packets processing capabilities in a virtualized cloud environment that enhance the scalability and high availability of the packets processing infrastructure. In certain embodiments disclosed herein, the VNICs functionality performed by network virtualization devices (NVDs) is offloaded from the NVDs to a fleet of computers, referred to as VNIC-as-a-Service System (or VNICaaS system). VNICaaS system is configured to provide Virtual Network Interface Cards (VNICs)-related functionality or service for multiple compute instances belonging to multiple tenants or customers of the CSPI. The VNICaaS system is capable of hosting multiple VNICs to process and transmit traffic in a distributed virtualized cloud networks environment. A single VNIC executed by the VNICaaS system can be used to process packets received from multiple compute instances.

Abstract: The present embodiments relate to a cloud shell extension framework. A cloud infrastructure (CI) service can invoke a cloud shell from a console to the CI. The service may request that context data to be added for use in a terminal session. The cloud shell extension can include a tool or script that can be used to obtain context data or sample code for preparing an environment specific to the service prior to the user interaction with the environment. The cloud shell extension can allow for a service to register an extension in the cloud shell framework, and their extension can be invoked when a client initiates the cloud shell with the service. The extension can allow for the invoked service to forward context data, such as environment variables, to be set in the terminal session for the client.

Abstract: Example embodiments facilitate efficient comparison operations of tree structures, resulting in comparison metrics (e.g., similarity or distance metrics or scores) used enhance software systems, such as search algorithms, code optimization software, enterprise database applications, and so on. Trees to be compared are converted into sets, i.e., serialized using a novel enumeration method. Metric functions can then be efficiently applied to the sets to facilitate the comparison operations. In an illustrative embodiment, subtrees of larger trees can be compared individually, pairwise, where the comparison results of the subtree comparisons can be selectively weighted and summed to yield an aggregated comparison metric that is tailored for a specific application or comparison priority.

Abstract: Techniques are disclosed for bootstrapping a secure data center using a cross domain system with a proxying sidecar node. The cross domain system can be implemented at the secure data center to provide one-way ingress and egress channels for network traffic to the target data center. The cross domain system is connected to a host data center and can receive configuration data from the host data center to configure the proxying sidecar node. The proxying sidecar node can request bootstrapping data from the host data center on demand, receive the requested bootstrapping data, and send the bootstrapping data to nodes in the secure data center to provision one or more services in the secure data center. The received bootstrapping data passes into the secure data center via the ingress channel.

Abstract: Techniques for providing, to a resource on a private network of a service provider, access to a resource on a private network of a customer. Service to customer (S2C) resources deployed on a cloud infrastructure to facilitate the access. Whereas IP address ranges may overlap between private networks and/or private IP addresses may be used in one or more of the private networks, the S2C resources enable the data exchange between the private networks. For example, the S2C resources translate between IP addresses such that data within each private network uses IP addresses that can be properly processed by the private network.

Abstract: Techniques are disclosed for bootstrapping a secure data center using a cross domain system and a cluster of edge devices. The cross domain system can be implemented at the secure data center to provide one-way ingress and egress channels for network traffic to the target data center. The cross domain system is connected to a host data center and can receive configuration data from the host data center to configure the cluster of edge devices. The cross domain system can include a cluster configuration node that can request bootstrapping data from the host data center on demand, receive the requested bootstrapping data, and use the bootstrapping data to generate a virtual bootstrapping environment on the cluster of edge devices.

Abstract: Techniques are disclosed for bootstrapping a secure data center using a cross domain system with a static sidecar node. The cross domain system can be implemented at the secure data center to provide one-way ingress and egress channels for network traffic to the target data center. The cross domain system is connected to a host data center and can receive configuration data from the host data center to configure the static sidecar node. The static sidecar node can receive bootstrapping data from the host data center and store the bootstrapping data. The bootstrapping data can include software resources for provisioning services in the secure data center. The received bootstrapping data passes into the secure data center via the ingress channel.

Abstract: Example embodiments facilitate prioritizing the recycling of computing resources, e.g., server-side computing systems and accompanying resources (e.g., non-volatile memory, accompanying firmware, data, etc.) leased by customers in a cloud-based computing environment, whereby computing resources (e.g., non-volatile memory) to be forensically analyzed/inspected, sanitized, and/or updated are prioritized for recycling based on estimates of when the computing resources are most likely to require recycling, e.g., via background sanitizing and updating. Computing resources that are likely to be recycled first are given priority over computing resources that are more likely to be recycled later. By prioritizing the recycling of computing resources according to embodiments discussed herein, other cloud-based computing resources that are used to implement computing resource recycling can be efficiently allocated and preserved.

Abstract: Techniques are disclosed for using a multi-tenant framework for microservices in a microservices-based application to handle tenant-specific circuit breaking. The microservices-based application can include at least one microservice that incorporates the multi-tenant framework. The multi-tenant framework includes software components configured to provide multi-tenant functionality for the microservice. A first microservice may receive an indication associated with a tenant that a second microservice is overloaded. The first microservice can then receive a request containing tenant context data at an interface. A first software component of the multi-tenant framework can extract the tenant context data from the request. The first microservice may determine whether the tenant context data corresponds to the tenant. The first microservice can then trigger a circuit breaker for the second microservice.

Abstract: A time series forecasting system is disclosed that obtains a time series forecast request requesting a forecast for a particular time point. The forecast request identifies a primary time series dataset for generating the requested forecast and a set of features related to the primary time series dataset. The system provides the primary time series dataset and the set of features to a model to be used for generating the forecast. The model computes a feature importance score for one or more features and selects a subset of features based on their feature importance scores. The model determines attention scores for a set of data points in the primary time series dataset based on the selected subset of features. The system predicts an actual forecast for the particular time point based on the attention scores and outputs the actual forecast and explanation information associated with the actual forecast.

Abstract: A system and technique for a Request Forwarder as for a computer network architecture is disclosed to provide selective access to one or more cloud services. In some implementations, a computer system may receive a request for access to a cloud service, the request including a container credential. The computer system may determine an identification of the container using the container credential. The computer system may verify that the container requesting access to the cloud service is authorized based at least in part on stored policies. Based at least in part on the determination that the container requesting access to the cloud service is authorized: receiving instance credential from a metadata service. The computer system may include the instance credential with the request. The computer system may send the request to the cloud service. In various examples, the Request Forwarder can be provided as a service.

Abstract: Log data that includes a plurality of log records is asynchronously processed to validate a configuration of each log record and data included in each log record. It is determined that one or more attributes of a particular subset of log records of the plurality of log records corresponds to one or more errors. Using the particular subset, one or more enriched log records are generated by augmenting each log record of the particular subset of log records with error information that indicates one or more categories corresponding to the one or more errors. A user interface is generated to facilitate correction of the one or more errors, the user interface comprising a plurality of interactive elements corresponding to a plurality of error metrics of different categories of errors, wherein the one or more categories of the one or more errors are included in the different categories of errors.

Abstract: The present disclosure relates to systems and methods that enhance log data processing by asynchronously processing log data for detecting log messages that include errors and enriching the log messages with annotations that identify the errors. More particularly, log data that includes log records can be received. The log data can be asynchronously processed to validate a configuration of each log record and to validate data included in each log record. An attribute of a particular log record can be determined to correspond to an error. A category of the error can be determined. An enriched log record can be generated by augmenting the particular log record with error information indicating the category. The enriched log record can be transmitted via a user interface to facilitate correction of the error.

Abstract: Techniques described herein relate to systems and methods of data storage, and more particularly to providing layering of file system functionality on an object interface. In certain embodiments, file system functionality may be layered on cloud object interfaces to provide cloud-based storage while allowing for functionality expected from a legacy applications. For instance, POSIX interfaces and semantics may be layered on cloud-based storage, while providing access to data in a manner consistent with file-based access with data organization in name hierarchies. Various embodiments also may provide for memory mapping of data so that memory map changes are reflected in persistent storage while ensuring consistency between memory map changes and writes. For example, by transforming a ZFS file system disk-based storage into ZFS cloud-based storage, the ZFS file system gains the elastic nature of cloud storage.

Abstract: The present disclosure relates to techniques for identifying out-of-domain utterances.

Abstract: Domain specific inlining for interpreters includes obtaining an interpreter source code including compiler directives. A host compilation is performed on the interpreter source code to obtain a compiled interpreter. Performing host compilation includes inlining code blocks based on the plurality of compiler directives. The compiled interpreter is outputted.

Abstract: A network virtualization device receives a storage request from a compute instance executed by a host machine separate from the network virtualization device, the storage request comprising data to be stored. Responsive to receiving the storage request, the network virtualization device divides the data to be stored into a set of one or more data stripes. For a first data stripe in the set of data stripes, the network virtualization device determines, from a plurality of extent servers, a first extent server to which the first data stripe is to be sent for storing the first data stripe, transmits the first data stripe to the first extent server, wherein the first extent server stores the first data stipe to a first physical storage device, and receives, from the first extent server, an acknowledgement confirming that the first data stripe was stored.