Abstract: A device receives source data, target data, external data, and a target task, and generates features of and differentiators between the source data and the target data. The device identifies a set of mappings between the source data and the target data based on the features and the differentiators, and determines different clusters of the source data based on the external data, the features, and the differentiators. The device generates, based on the external data, a set of artificial intelligence (AI) models as candidates to perform the target task, and generates a performance measure for the set of AI models based on the features, the differentiators, and the external data. The device refines the set of mappings, and identifies an AI model, from the set of AI models, to perform the target task based on the different clusters of the source data and based on the performance measure.

Abstract: A method is disclosed for dynamic multitasking in a storage system, the storage system including a first storage server configured to execute a first I/O service process and one or more second storage servers, the method comprising: detecting a first event for triggering a context switch; transmitting to each of the second storage servers an instruction to stop transmitting internal I/O requests to the first I/O service process, the instruction including an identifier corresponding to the first I/O service process, the identifier being arranged to distinguish the first I/O service process from other first I/O service processes that are executed by the first storage server concurrently with the first I/O service process; deactivating the first I/O service process by pausing a frontend of the first I/O service process, and pausing one or more I/O providers of the first I/O service process; and executing a first context switch between the first I/O service process and a second process.

Abstract: A set of computing nodes may receive a corresponding set of heartbeat messages that originated at the set of computing nodes. The set of heartbeat messages may relate to selecting, among the set of computing nodes, a leader computing node to process a set of jobs. State information included in the heartbeat messages may be provided to a leader election algorithm that outputs information indicating one or more computing nodes that are most qualified to process the set of jobs based on processing capabilities of the computing nodes and processing constraints associated with the set of jobs. The computing node may select itself as the leader computing node to process the set of jobs based on determining, from the information output by the leader election algorithm, that the computing node is most qualified to process the set of jobs and no other computing nodes are processing the set of jobs.

Abstract: An apparatus includes a memory and a hardware processor. The memory stores one or more scheduling histories. The processor receives a first list from a first client device and a second list from a second client device. The first list includes a first job and the second list includes a second job. The processor calculates a first resource requirement to execute the first job and a second resource requirement to execute the second job. The processor determines that a conflict exists between the first job and the second job and in response, communicates messages to the first and second client devices indicating the conflict and proposed times for executing the jobs. The processor determines based on responses to the messages, execution times for the first and second jobs, assigns permission identifiers to the first and second jobs, and communicates a schedule.

Abstract: Disclosed is a computer system for allocating a resource, such as a computing resource. In an aspect, there is provided a method for allocating a resource associated with a root node. The method includes: authenticating a network node by determining that the network node is child node, the child node being a node associated with a child in a hierarchical model; receiving, via a communication module and from the child node, a signal representing resource distribution data, the resource distribution data indicating a proportional distribution of resources among descendants of the child; and in response to receiving the signal representing resource distribution data, allocating the resource to the descendants of the child by: i) determining a total amount of resources allocated to the child; and ii) based on the resource distribution data, allocating respective portions from the total amount of resources allocated to the child to descendants of that child.

Abstract: Complex workflows are composed of a plurality of idempotent actions. During execution of the complex workflows, a computing system accesses the plurality of idempotent actions and determines whether corresponding guard conditions for triggering processing of the idempotent action are satisfied. When satisfied, a lock is taken on one or more resources used for executing the idempotent code of the idempotent action and execution of the idempotent code is initiated. Thereafter, upon successful execution of the idempotent code, the corresponding record is updated to reflect execution of the idempotent action and the lock is released. When execution of the idempotent action is unsuccessful, an exception is logged and the lock is released.

Abstract: A system for cluster resource allocation includes an interface and a processor. The interface is configured to receive a process and input data. The processor is configured to determine an estimate for resources required for the process to process the input data; determine existing available resources in a cluster for running the process; determine whether the existing available resources are sufficient for running the process; in the event it is determined that the existing available resources are not sufficient for running the process, indicate to add new resources; determine an allocated share of resources in the cluster for running the process; and cause execution of the process using the share of resources.

Abstract: Scheduling job request submitted by multiple tenants in a manner that honors multiple software license agreements for the multiple tenants. A queue persistently stores job requests that await scheduling. A job state tracking component persistently tracks a state of the job requests, and perhaps provides job requests into the queue. A software license agreement enforcer reviews the job requests in the queue, selects one or more job requests should be scheduled next based on the license agreements, and provide the selected job requests to a resource manager. A subscriber/publisher pool may be used to the various components to communicate. This decouples the communication from being a simple one-to-one correspondence, but instead allows communication from a component of one type to a component of the other type, whichever instance of those components happens to be operating.

Abstract: A distributed node intra-group task scheduling method includes: step 101) collecting resource state information and history service information about each node in a group; step 102) receiving a service request command and parsing the request command to obtain a request task; step 103) according to the resource state information and history service information about each node, calculating the resource utilization rate increment after each node in the group loads a task and the occupation condition of each kind of resource in the group; and according to a principle of making the resource utilization rate increment of each node in the group as low as possible and the occupation of each kind of resource in the group as balanced as possible, selecting service execution nodes in the group, and providing the request task by the selected service execution nodes.

Abstract: Systems and methods of managing computational resources are provided. In one exemplary embodiment, a method by a controller (305, 407, 500, 600, 700, 1101) for managing computational resources may include dynamically distributing (801) computational resource shares among sequential services that are mapped to one or more processors (303, 403). Further, each sequential service corresponds to an execution step of a remote application (307, 409). Also, a service chain (313-315, 413-415) comprises at least one sequential service. The dynamical distribution is based on estimated and predetermined tail latencies and average execution times of each sequential service in the service chain as well as the service chain such that the latencies are met. In addition, the one or more service chains are executed contemporaneously.

Abstract: Disclosed herein are embodiments for providing finite state machine driven workflows. In an embodiment, a workflow template is defined for a type of task. The workflow template may represent a finite state machine. The workflow template may be linked to an external party and an asset type, which may be stored in a workflow database. An asset may be received from the external party including an external party attribute identifying the external party, an asset type attribute, and an owner attribute. The owner attribute may be associated with an application end user. A determination may be made whether the external party attribute and the asset type attribute of the asset match the external party and the asset type linked to the workflow template. If a match is determined, instances of the task and the one or more actions of the workflow template may be created.

Abstract: A terminal application process management method includes: detecting a first-type terminal application, the first-type terminal application being a preset terminal application that needs to be accelerated but has not been started yet; obtaining attribute information of a second-type terminal application if the first-type terminal application is trigger-started, the second-type terminal application being another currently-running terminal application; and terminating a process of the second-type terminal application according to the attribute information of the second-type terminal application until there is sufficient resource for starting the first-type terminal application. Embodiments of the present invention further provide a terminal application process management apparatus.

Abstract: Disclosed are various examples of replication management for hyper-converged infrastructures. Virtual machine groups are generated using k-means grouping based on a process list of a respective virtual machine of a plurality of virtual machines within a hyper-converged infrastructure. Virtual machines in a respective group are analyzed to determine a first set of resources. A property graph that includes configuration data including a storage resource configuration and a network resource configuration is generated for the first set of resources of the respective virtual machine group. A second set of resources is configured within a second workload domain using the storage resource configuration and the network resource configuration.

Abstract: Systems and methods for monitoring one or more servant regions of an application server. A processor in electronic communication with the application server identifies one or more servant regions of the application server; requests to the application server to receive one or more performance metrics for each of the one or more servant regions, the performance metrics included in a virtual resource statistics module for collecting and reporting available resource data; receives from the application server first information including the one or more performance metrics for each of the one or more servant regions; extracts from the first information the one or more performance metrics for each of the one or more servant regions of the application server; and determines, based on the one or more performance metrics, whether to generate an alert message.

Abstract: A system according one exemplary embodiment may receive instructions regarding a callee function that include a description of properties associated with arguments for the callee function, create a persistent representation of the callee function based on the received description of the callee function, receive instructions from a source describing a set of properties for a data collection, create a persistent representation of the data collection based on the set of properties for the data collection, receive an updated property for the data collection, and in response to receiving the updated property for the data collection: update the persistent representation of the callee function by associating the persistent representation of the callee function with the updated property of the data collection and update the persistent representation of the data collection by associating the persistent representation the data collection with the updated property of the data collection.

Abstract: Systems and methods for sending an in-application message to a user of a software application makes use of templated messages. User-specific information is added to the template to create the message delivered to the user. The user-specific information is added to the template to create the message immediately before the message is to be delivered to the user to ensure the user-specific information is as up-to-date as possible.

Abstract: Multiple processes of an application (or multiple applications capable of sharing private data) of a mobile device can communicate in a process-safe, secure, and efficient manner via a shared private file system. A first process can produce shared data for a second process to consume via the file system. For example, when the shared data changes, the first process can write a new file to the private file system and rename the new file to a specified file name. When the second process needs the shared data, the second process can evaluate certain metadata of the new file, such as a timestamp or a field comprising a monotonically increasing number. If a stored metadata value associated with the specified file name is less than the metadata value of the new file, the second process can safely retrieve the shared data from the new file.

Abstract: Some embodiments relate to assigning individual log messages to clusters. An initial cluster assignment may be performed by applying a hash function to one or more non-variable components of the message to generate an initial cluster identifier. Subsequently, clustering may be further refined (e.g., by determining whether to merge clusters based on similarity values). An interface can present a representative message of each cluster and indicate which portions of the message correspond to a variable component. Particular inputs detected at the input corresponding to one of these components can cause other values for the component to be presented. For a given cluster, timestamps of assigned messages can be used to generate a time series, which can facilitate grouping of clusters (with similar or complementary shapes) and/or triggering alerts (with a condition corresponding to a temporal aspect).

Abstract: Flow control of messages to/from clients and services in a remote access system. Within the remote access system a layered architecture may be provided that includes a client layer at a client device, a service layer associated with a service, and a server layer that is a communications proxy for messages sent between the client device and the service. The service layer may have a receive queue for messages received from the client device and a send queue of messages destined for the client device. The server layer may include an outbound queue of messages destined for the client device. Full duplex communications between the client layer and the server layer may be implemented using a service protocol in a first channel for communication between the client layer and the server layer and a second channel for communication between the server layer and the client layer.

Abstract: An electronic device includes one or more integrated circuits, a debugging translation block, and a bus connected to the one or more integrated circuits and the debugging translation block, the bus configured to provide a connection to one or more external devices, wherein the debugging translation block is configured to receive debugging commands from a testing host device via the bus, convert the debugging commands into debugging input data, and provide the debugging input data to a debugging state machine of a first integrated circuit of the one or more integrated circuits.

Abstract: An encoder includes: a reading device that reads respective electric signals from two incremental patterns respectively having graduation array pitches different from each other; a control device that calculates a measurement value, based on the electric signals; and an output device that outputs the measurement value. The control device includes: an absolute position synthesis unit that synthesizes two electric signals to generate a synthesized absolute position; a detection unit that detects two relative positions from the two electric signals; a position calculation unit that performs an arithmetic operation between the relative positions and the synthesized absolute position to calculate a calculated absolute position; an absolute position comparison unit that compares the calculated absolute position with the synthesized absolute position; and a relative position comparison unit that compares the two relative positions with each other.

Abstract: A method, a computer-readable medium, and an apparatus are provided. The apparatus may be an embedded-system device. The embedded-system device determines a respective operational state of each of one or more processes of the embedded-system device. The embedded-system device stores the respective operational state of each of the one or more processes at a memory location in a respective memory area for the each process in a memory of the embedded-system device. The embedded-system device stores the memory locations associated with the one or more processes in a register in the memory. The embedded-system device obtains, from the register, a memory location of at least one process of the one or more processes. The embedded-system device obtains, based on the memory location of the at least one process, the stored operational state of the at least one process from the respective memory area for the at least one process.

Abstract: A device may receive information related to an operation of the device during each of multiple time intervals. The device may store, for each time interval of the multiple time intervals, the information in a respective slot of a circular buffer that includes multiple slots. The circular buffer may be used to store a historical record of the information in one or more of the multiple slots. The historical record may be provided from the circular buffer during a dump of the device. The device may provide the historical record during the dump of the device based on storing the historical record of the information in the one or more of the multiple slots.

Abstract: A method and system for server fault analysis are described. A server configuration analyzer retrieves event logs collected from a server with a fault such as a misconfiguration. These event logs can then be analyzed using metrics to identify one or more messages in the logs that indicate a cause of the fault on the server. Once the messages are identified, the server configuration analyzer retrieves remedial commands that may be used to correct the fault from a database that stores remedial commands correlated with event log messages. The server configuration analyzer can then present the messages and remedial commands to a user to assist with identifying the cause of the fault and correcting it.

Abstract: A method is provided for clustering events. Messages are received at an extraction engine from managed infrastructure that includes managed infrastructure physical hardware that supports the flow and processing of information. Events are produced that relate to the managed infrastructure. The events are converted into words and subsets used to group the events that relate to failures or errors in the managed infrastructure, including the managed infrastructure physical hardware. One or more common characteristics of events are determined and clusters of events are produced relating to the failure or errors in the managed infrastructure. A source address is used for each event as is a graph topology of the managed infrastructure to assign a graph coordinate to the event. Membership in a cluster indicates a common factor of the events that is a failure or an actionable problem in the physical hardware managed infrastructure directed to supporting the flow and processing of information.

Abstract: A system includes: at least one processor running instances of two or more services. Instances of a first service are dependent upon instances of a second service. A fault-injection manager, based on data in a service directory indicative of dependencies of services of the system, determines a fault-inducing condition to inject into the system and injects the fault-inducing condition into the system. A health monitoring manager detects failures in the system and identifies the first service as vulnerable to the fault-inducing condition when the fault-inducing condition causes an instance of the second service to fail.

Abstract: Embodiments of the present disclosure provide an information processing method and apparatus. The method comprises: sending a first message to a first equipment, wherein the first message is used for requesting user data of a user equipment; receiving a second message sent by the first equipment according to the first message, wherein the second message carries the user data of the user equipment and a network access failure transmission indication; and when determining a network access failure of the user equipment, sending a network access failure cause value of the user equipment to a storing user subscribed information entity or a failure cause processing entity according to the network access failure transmission indication. The information processing method and apparatus provided by the embodiments of the present disclosure may be used for achieving a failure diagnosis of the user equipment.

Abstract: An automated agent for the causal mapping of complex environments. Specifically, a disclosed method and system entails the application of statistical tools, or causality tests, to measure the strength and direction of causal relations between two or more environment components. Further, the execution of the causality tests may be an offline process that may be triggered periodically to account for changes or updates to an environment over time.

Abstract: A point cloud filter method and apparatus is provided for use in point cloud rendering from real-time point cloud data collection from a variety of sensor types is provided that delivers enhanced performance including reducing processing requirements, limiting local memory consumption and optimizing overall data visualization.

Abstract: A data storage device includes a memory, a first module, and a second module. The first module is configured to sense data stored at the memory to generate a first set of soft bits having a first number of bits. The second module is configured to perform an operation using the first set of soft bits to generate a second set of soft bits having a second number of bits that is less than the first number of bits. In an illustrative implementation, the second set of soft bits is used in connection with a three-stage decoding process to decode a set of hard bits that represents the data.

Abstract: A cache controller id disclosed, The cache controller includes circuitry to receive a request to access data in a target location of a last level cache of a processor on a processor package, identify an in-field failure in the target location of the last level cache, perform, in response to the identification of the in-field failure, an in-field repair, including circuitry to write in-field repair information to a non-volatile memory on the processor package and external to the processor, the non-volatile memory including circuitry to store in-field repair information. Systems and methods are also disclosed.

Abstract: Systems and methods are disclosed for host assisted error recovery at a data storage device. In an example embodiment, an apparatus comprises a data storage drive including an interface to communicate with a host device, a nonvolatile solid state memory, and a processor. The processor is configured to calculate parity data based on a plurality of pages from the nonvolatile solid state memory, and provide, in response to a request for the parity data received from the host device via the interface, the parity data to the host device with an indication to store the parity data to a nonvolatile storage medium.

Abstract: A system and method pertains to operating non-volatile memory systems. Technology disclosed herein efficiently uses memory available in non-volatile storage devices in a non-volatile memory system. In some aspects, non-volatile storage devices enforce a redundancy coding stripe across the non-volatile storage devices formed from chunks of data having internal addresses assigned in a coordinated scheme across the storage devices. In some aspects, non-volatile storage devices enforce a redundancy coding stripe across the non-volatile storage devices at the same internal addresses in the respective non-volatile storage devices.

Abstract: An error checking and correcting (ECC) decoding method and apparatus are provided. A decoding circuit decodes a codeword using (or without using) reference information, wherein when the decoding circuit fails to decode a first codeword, the decoding circuit decodes a second codeword to produce decoded data. The decoding circuit checks whether a change has occurred from each codeword bit of the second codeword to a corresponding bit of the decoded data. In accordance with a bit position of the changed corresponding bit, the decoding circuit correspondingly changes the first codeword to a modified codeword, and/or correspondingly changes the reference information to modified information. The decoding circuit performs the ECC decoding again on the modified codeword (or the first codeword) using (or without using) the modified information.

Abstract: A data storage device includes a flash memory and a controller. The flash memory includes a plurality of planes, and each of the planes includes a plurality of blocks. Each of the blocks includes a plurality of pages. The size of each page is N K-bytes, wherein N is a positive integer greater than 1. The controller is coupled to the flash memory to calculate the ECC bit number of each page using a detection unit of 1 Kbyte. The controller statistically calculates the number of detection units of the pages corresponding to different values of the ECC bit number in order to determine whether each plane of the flash memory is normal or not.

Abstract: A processing device has a memory including a plurality of storage areas each storing a setting value and the correction frequency change unit. The correction frequency change unit is configured to decrease, when a frequency of correction by the error correction unit for a storage area that stores one of the setting values is changed to a high frequency, a frequency of correction by the error correction processing for one of the storage areas other than the storage area whose correction frequency is changed to a high frequency, and to increase, when a frequency of correction by the error correction unit for a storage area that stores one of the setting values is changed to a low frequency, a frequency of correction by the error correction processing for one of the storage areas other than the storage area whose correction frequency is changed to a low frequency.

Abstract: Techniques are disclosed for provided automatic fault tolerance in data storage systems and other computer systems managing access to shared computing resources using the construct of a resource instance manager together with an architecture for designing and implementing task-oriented state-based commands for accessing the shared computing resources. A set of resource instance managers respond to all commands requesting access to a set of respective shared resources. Access to each shared resource is managed by a unique resource instance manager for that resource which maintains a consistent state for that shared resource. When commands are designed appropriately, the system can be restored using a system state history maintained transparently by a system-provided application program interface.

Abstract: Apparatus adapted for exascale computers are disclosed. The apparatus includes, but is not limited to at least one of: a system, data processor chip (DPC), Landing module (LM), chips including LM, anticipator chips, simultaneous multi-processor (SMP) cores, SMP channel (SMPC) cores, channels, bundles of channels, printed circuit boards (PCB) including bundles, floating point adders, accumulation managers, QUAD Link Anticipating Memory (QUADLAM), communication networks extended by coupling links of QUADLAM, log 2 calculators, exp2 calculators, log ALU, Non-Linear Accelerator (NLA), and stairways. Methods of algorithm and program development, verification and debugging are also disclosed. Collectively, embodiments of these elements disclose a class of supercomputers that obsolete Amdahl's Law, providing cabinets of petaflop performance and systems that may meet or exceed an exaflop of performance for Block LU Decomposition (Linpack).

Abstract: In a block-based backup system, a technique allows indexing changed files between backups so that files not changed are not re-indexed. Block identifiers specifying backed-up blocks are received. A file table record is accessed to determine a location of a file table zone. Records in the file table store last modified timestamps for files associated with the records. The file table zone location is compared with the block identifiers to select a subset of records. Each record in the subset is associated with a block from the file table zone that corresponds to a block identified by the block identifiers. The subset is parsed to extract timestamps. The timestamps are compared with a next-to-last backup date. If the associated file was modified after the next-to-last backup date, a determination is made that the file blocks were backed up in the last backup and the file should be indexed.

Abstract: Example embodiments relate to asset browsing and restoration over a network using on demand staging. A method may include accessing, by a storage access system, a first portion of a compound asset in a storage system and restoring it as a staged asset in a staging area of the storage access system. The accessing and restoring of the first portion may be performed on demand in response to an indication from a client device to browse a portion of the compound asset. The method may further include accessing the internal structure of the staged asset to generate an asset directory. The asset directory may indicate discrete items within the compound asset. The method may further include providing the asset directory to the client device over a network, and receiving an indication from the client device of at least one of the discrete items to restore to the client device.

Abstract: Determining the bandwidth required for data replication for disaster recovery. Given a specified recovery point objective (RPO), bandwidth requisitioning and usage is determined according to meet the RPO based on observed factors.

Abstract: Utilizing an incremental backup in a decremental backup system. In one example embodiment, a method for utilizing an incremental backup in a decremental backup system includes creating a base backup that includes all original allocated blocks in a source storage at a first point in time. Next, an incremental backup is created that includes allocated blocks in the source storage that were newly allocated or were changed between the first point in time and a second point in time. The changed allocated blocks in the incremental backup are identified. During a third time period, a decremental backup is created and the incremental backup is ingested into the base backup by copying, into the decremental backup, original allocated blocks from the base backup that correspond to the changed allocated blocks in the incremental backup. Lastly, the newly-allocated blocks and the changed allocated blocks from the incremental backup are added to the base backup.

Abstract: Rather than relying on pre-defined scheduling of secondary copy operations such as backup jobs, the illustrative opportunistic approach initiates secondary copy operations based on changing operational conditions in a storage management system. An adaptive backup readiness score is based on a number of backup-readiness operational factors. An illustrative enhanced data agent which is associated with the target database application (or other executable component) may monitor the operational factors and determine the backup readiness score based on weights assigned to the respective operational factors. The enhanced data agent may evaluate recent backup jobs to determine which of the operational factors that contributed to the backup readiness score may have been most relevant.

Abstract: Embodiments described herein provide for a system and associated logic to enable browsing of a federated backup having a plurality of portions, each portion separately generated and indexed by a plurality of nodes. In one embodiment, the system performs a federated backup by generating a unique identifier on a primary node and providing the unique identifier to each secondary node tasked with generating the plurality of portions. The secondary nodes associate the unique identifier with each of the plurality of portions. The unique identifier can then be used by a recovery UI to generate and display a consolidated view of the plurality of portions.

Abstract: A method for contact information backup and recovery comprising receiving, by a contact information backup and recovery system, subscriber information, storing the subscriber information in a contacts data store, receiving, by the contact information backup and recovery system, contacts information associated with the subscriber information, storing the contacts information in the contacts data store, wherein the contacts information is stored in relation to the subscriber information, receiving, by the contact information backup and recovery system, a contacts recovery request, and performing, by the contact information backup and recovery system, a contacts recovery operation, comprising acquiring target contact information from the stored contacts information in the contacts data store, performing a telephonic communication using the target contact information, and sending the stored contacts information to a remote storage device.

Abstract: Recovery of a database system can be initiated by taking the database system offline. Thereafter, recovery operations specified by a redo log of the database system are replayed. During the replaying, updates to pages implicated by the recovery operations are blocked. In parallel, modified pages are flushed to physical disk storage. The database is later brought online after all of the recovery operations are replayed.

Abstract: Recovery points can be used for replicating a virtual machine and reverting the virtual machine to a different state. A filter driver can monitor and capture input/output commands between a virtual machine and a virtual machine disk. The captured input/output commands can be used to create a recovery point. The recovery point can be associated with a bitmap that may be used to identify data blocks that have been modified between two versions of the virtual machine. Using this bitmap, a virtual machine may be reverted or restored to a different state by replacing modified data blocks and without replacing the entire virtual machine disk.

Abstract: A method for economizing computing resources and verifying an integrity of parameters of a neural network by inserting test pattern into a background area of an input image is provided for fault tolerance, fluctuation robustness in extreme situations, functional safety on the neural network, and an annotation cost reduction. The method includes: a computing device (a) generating t-th background prediction information of a t-th image by referring to information on each of a (t?2)-th image and a (t?1)-th image; (b) inserting the test pattern into the t-th image by referring to the t-th background prediction information, to thereby generate an input for verification; (c) generating an output for verification from the input for verification; and (d) determining the integrity of the neural network by referring to the output for verification and an output for reference. According to the method, a data compression and a computation reduction are achieved.

Abstract: A system with distributed monitoring agents include a state storage, a plurality of worker agents, a first processor, and a second processor. A job is executed using a worker agent of the plurality of worker agents. The first processor is configured to execute a first monitor to monitor the job and to restart the job using job state data stored in the state storage in the event that the job fails to successfully complete. The second processor is configured to execute a second monitor to monitor the first monitor and to restart the first monitor using first monitor state data stored in the state storage in the event that the first monitor crashes.

Abstract: An apparatus in one embodiment comprises a storage system including a plurality of storage nodes each associated with one or more storage devices. The storage system is configured to provide at least one virtual volume distributed over the storage nodes for utilization by a plurality of host devices. The storage nodes comprise respective input-output sequencers for handling input-output operations directed to the distributed virtual volume by the host devices. The input-output sequencers of the respective storage nodes are configured to interact with one another to collectively form a distributed input-output sequencer that ensures consistency across the storage nodes for the distributed virtual volume when accessed by different ones of the host devices. The storage system may comprise a multi-site storage system, with the plurality of storage nodes comprising at least a first storage node cluster at a first site and a second storage node cluster at a second site.