Abstract: In a multi-node system, each node includes tiles. Each tile includes a cache controller, a local cache, and a snoop filter cache (SFC). The cache controller responsive to a memory access request by the tile checks the local cache to determine whether the data associated with the request has been cached by the local cache of the tile. The cached data from the local cache is returned responsive to a cache-hit. The SFC is checked to determine whether any other tile of a remote node has cached the data associated with the memory access request. If it is determined that the data has been cached by another tile of a remote node and if there is a cache-miss by the local cache, then the memory access request is transmitted to the global coherency unit (GCU) and the snoop filter to fetch the cached data. Otherwise an interconnected memory is accessed.

Abstract: A system includes a first counter configured to increment or decrement in response to a triggering event. The first counter is sized to overflow. The system also includes a second counter configured to increment or decrement in response to a triggering event. The first counter and the second counter are merged to form a third counter in response to detecting an overflow triggering event for the first counter. A merge bit indicative of whether the first counter and the second counter are merged changes value in response to merging the first counter and the second counter.

Abstract: In a multi-node system, each node includes tiles. Each tile includes a cache controller, a local cache, and a snoop filter cache (SFC). The cache controller responsive to a memory access request by the tile checks the local cache to determine whether the data associated with the request has been cached by the local cache of the tile. The cached data from the local cache is returned responsive to a cache-hit. The SFC is checked to determine whether any other tile of a remote node has cached the data associated with the memory access request. If it is determined that the data has been cached by another tile of a remote node and if there is a cache-miss by the local cache, then the memory access request is transmitted to the global coherency unit (GCU) and the snoop filter to fetch the cached data. Otherwise an interconnected memory is accessed.

Abstract: Systems and methods for using a thin arbiter node as a quorum decision maker in a distributed file system. An example method comprises receiving a write request with respect to a file, identifying a first data storage unit storing a first replica of the file and a second data storage unit storing a second replica of the file, determining that the first data storage unit is unavailable and that the second data storage unit is available, and in response to determining, using metadata residing on a thin arbiter node, that the second data storage unit contains at least on stale file replica, failing the write request to avoid split brain in the distributed file system.

Abstract: A request to perform a write operation on a file stored in a distributed file system may be received. A determination may be made as to whether a quorum of servers of the distributed file system is satisfied. The servers of the quorum may be used to perform the write operation or to record the write operation. The write operation may be performed on the file in view of determining that the quorum has been satisfied.

Abstract: A first storage server of the file system receives a request to perform an operation on the data file. The operation is to be performed on a first replica of the data file stored at the first storage server and a second replica of the data file stored at a second storage server of the file system. The first storage server configures first metadata associated with a first index file to indicate that the operation is to be performed on a first portion of the first replica. The first storage server determines the second replica is an outdated state indicating that the operation on the second replica has not been performed by the second storage server. In response to the second replica being in the outdated state, updating a first portion of the second replica identified in view of the first metadata and corresponding to the first portion of the first replica.

Abstract: Delayed asynchronous file replication in a distributed file system is described. A first server receives requests to perform a updates to a first replica of a file. The first server sends, to a second server, an outcast instruction to set a second replica of the file to an out-of-date state. The first server performs the updates to the first replica without sending the updates to the second server. The first server receives a request for updates for the second replica from the second server. The first server transmits the updates to the second server to facilitate a heal operation of the second replica, where completion of the heal operation causes the out-of-date state of the second replica to be changed to an up-to-date state.

Abstract: Systems and methods for migrating locking data for a file system object within a distributed file system. An example method may comprise: initiating a transfer of locking data from a first node of a distributed file system to a second node of the distributed file system, wherein the locking data is associated with a file system object and comprises a connection identifier indicating a connection between a client requesting a lock and the first node; constructing a modified connection identifier that indicates a connection between the client and the second node; and updating the second node to include the locking data with the modified connection identifier.

Abstract: Systems and methods for using a thin arbiter node as a quorum decision maker in a distributed file system. An example method comprises receiving a write request with respect to a file, identifying a first data storage unit storing a first replica of the file and a second data storage unit storing a second replica of the file, determining that the first data storage unit is unavailable and that the second data storage unit is available, and in response to determining, using metadata residing on a thin arbiter node, that the second data storage unit contains at least on stale file replica, failing the write request to avoid split brain in the distributed file system.

Abstract: A request to perform a write operation on a file stored in a distributed file system may be received. A determination may be made as to whether a quorum of servers of the distributed file system is satisfied. The servers of the quorum may be used to perform the write operation or to record the write operation. The write operation may be performed on the file in view of determining that the quorum has been satisfied.

Abstract: A request to perform a write operation on a file stored in a distributed file system that includes a first and second storage server and an arbiter system may be received. An identification of whether whether one of the first or second storage servers is available to perform the write operation and that the other of the first or second storage servers is not available to perform the write operation may be performed. An identification of whether the arbiter system is available to record the write operation may be performed. In response to identifying that one of the storage servers and the arbiter system are available and that the other of the storage servers is unavailable, the write operation may be performed on the file in view of write operation data that indicates whether the one of the storage servers is consistent with the arbiter system.

Abstract: An outcast index in a distributed file system is described. A first server can receive an indication that a first replica stored on the first server is to be modified in view of a second replica stored on a second server. The first replica and the second replica are replicas of a same file. The first server updates metadata associated with the first replica to indicate an outcast state of the first replica. The first server receives an indication that the modification of the first replica is complete. The first server updates the metadata associated with the first replica to remove the outcast state of the first replica.

Abstract: Systems and methods for migrating locking data for a file system object within a distributed file system. An example method may comprise: initiating a transfer of locking data from a first node of a distributed file system to a second node of the distributed file system, wherein the locking data is associated with a file system object and comprises a connection identifier indicating a connection between a client requesting a lock and the first node; constructing a modified connection identifier that indicates a connection between the client and the second node; and updating the second node to include the locking data with the modified connection identifier.

Abstract: A granular data self-healing manager in a file system is described. A first storage server of the file system receives a request to perform an operation on the data file. The operation is to be performed on a first replica of the data file stored at the first storage server and a second replica of the data file stored at a second storage server of the file system. The first storage server configures first metadata associated with a first index file to indicate that the operation is to be performed on a first portion of the first replica. The first storage server determines the second replica is an outdated state indicating that the operation on the second replica has not been performed by the second storage server. In response to the second replica being in the outdated state, updating a first portion of the second replica identified in view of the first metadata and corresponding to the first portion of the first replica.

Abstract: A request to perform a write operation on a file stored in a distributed file system that includes a first and second storage server and an arbiter system may be received. An identification of whether whether one of the first or second storage servers is available to perform the write operation and that the other of the first or second storage servers is not available to perform the write operation may be performed. An identification of whether the arbiter system is available to record the write operation may be performed. In response to identifying that one of the storage servers and the arbiter system are available and that the other of the storage servers is unavailable, the write operation may be performed on the file in view of write operation data that indicates whether the one of the storage servers is consistent with the arbiter system.

Abstract: A server computer system performs a first set of operations for a first transaction. The first transaction pertaining to data stored in a file system. The server computer system delays a second set of operations for the first transaction and identifies a second transaction pertaining to the data. In response to identifying the second transaction, the server computer system cancels the second set of operations for the first transaction, and cancels a first set of operations for the second transaction.

Abstract: Techniques for pro-active self-healing in a distributed file system are disclosure herein. In accordance with one embodiment, a method is provided. The method comprises prior to detecting an access request by a client application to an image on a storage server, identifying, by a self-healing daemon executed by a processing device, a first region of the image comprising stale data. A partial lock on the image is acquired. The partial lock prevents access to the first region of the image. Responsive to acquiring the partial lock, the self-healing daemon provides access to a second region of the image file comprising data other than the stale data.

Abstract: Techniques for pro-active self-healing in a distributed file system are disclosure herein. In accordance with one embodiment, a method is provided. The method comprises prior to detecting an access request by a client application to an image on a storage server, identifying, by a self-healing daemon executed by a processing device, a first region of the image comprising stale data. A partial lock on the image is acquired. The partial lock prevents access to the first region of the image. Responsive to acquiring the partial lock, the self-healing daemon provides access to a second region of the image file comprising data other than the stale data.

Abstract: A data archiving device can be connected to a host device, and can include a version control system (VCS) unit configured to store archive data in a storage unit in response to the data archiving device detecting modifications to a host file of the host device. The archive data corresponds to the modifications and is processable to reconstruct previous versions of the host file. The VCS unit can be located in firmware that prevents overwriting and erasure operations in the storage unit by a user. The data archiving device can retrieve, upon receiving a request for a previous version of the host file, relevant archive data associated with the previous version of the host file. The data archiving device can forward the relevant archive data to the management device to enable the management device to reconstruct the previous version of the host file.

Abstract: A first storage server identifies a second storage server connecting to a cluster of storage servers in a file system. The first storage server locates a file in a replication directory hierarchy in the second storage server. The file has not been accessed by a client application. The first storage server determines, prior to the file being accessed by the client application, that the file on the second storage server has stale data and overwrites the stale data in the file on the second storage server with current data.