# US 10,455,045 B2 · Automatic data replica manager in distributed caching and data processing systems > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 10,455,045 B2 - Document ID: US10455045B2 - Status: Granted - Kind: B2 - Application number: 15/408,328 - Filed: 2017-01-17 - Published: 2018-03-08 - Granted: 2019-10-22 - Earliest claimed: 2016-09-06 - Assignees: Samsung Electronics Co., Ltd. - Applicants: Samsung Electronics Co., Ltd. - Inventors: Zhengyu Yang; Jiayin Wang; Thomas David Evans - Web page: https://patentgrep.com/patent/US10455045B2 ## CPC - H04L 67/568: Storing data temporarily at an intermediate stage, e.g. caching - G06F 9/00: Arrangements for program control, e.g. control units (program control for peripheral devices G06F13/10) - H04L 41/0668: by dynamic selection of recovery network elements, e.g. replacement by the most appropriate element after failure - H04L 41/5009: Determining service level performance parameters or violations of service level contracts, e.g. violations of agreed response time or mean time between failures [MTBF] - H04L 41/5022: by giving priorities, e.g. assigning classes of service - H04L 41/5025: by proactively reacting to service quality change, e.g. by reconfiguration after service quality degradation or upgrade - H04L 43/0852: Delays - H04L 43/0888: Throughput - H04L 43/16: Threshold monitoring - H04L 67/1001: for accessing one among a plurality of replicated servers - H04L 67/1095: Replication or mirroring of data, e.g. scheduling or transport for data synchronisation between network nodes - H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M) - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract A method of data storage includes determining a latency distance from a primary node to each of two or more replica nodes, choosing a preferred replica node of the two or more replica nodes based on the determined latency distances, and write-caching data into the preferred replica node. ## Claims ### Claim 1 A method for enabling fault tolerance of a datacenter cluster caching system having one or more replica nodes in the same tier, the method comprising: detecting a failure of the datacenter cluster; performing load balancing among nodes of the datacenter cluster in accordance with the detected failure by moving data among storage tiers of the nodes; and migrating replica data from a first replica node in the datacenter cluster to a second replica node in the datacenter cluster by: writing new incoming data on the second replica node; and maintaining unchanged cached data on the first replica node, wherein the load balancing is performed without pausing a subsystem corresponding to the detected failure, and without copying an entirety of a dataset of a node of the datacenter cluster corresponding to the detected failure. ### Claim 2 (depends on 1) The method of claim 1, further comprising: evicting victim data from a solid-state drive of a primary node; receiving the new incoming data at the primary node; writing the new incoming data to a hard disk drive of the primary node; and removing the victim data from a solid-state drive of the first replica node. ### Claim 3 (depends on 1) The method of claim 1, wherein the detected failure comprises a virtual machine crash on a primary node of the datacenter cluster, the method further comprising: writing dirty data from a solid-state drive of the primary node to a hard disk drive of the primary node; maintaining the dirty data in a solid-state drive of a replica node; flagging the dirty data in the solid-state drive of the replica node as nondirty; restarting the crashed virtual machine on the primary node; and forwarding incoming I/O requests to the solid-state drive of the primary node and to the solid-state drive of the replica node. ### Claim 4 (depends on 1) The method of claim 1, wherein the detected failure comprises a cache device failure on a primary node of the datacenter cluster such that information cannot be written to the primary node, the method further comprising: writing dirty data from a solid-state drive of a replica node of the datacenter cluster to a hard disk drive of the primary node; maintaining the dirty data in the solid-state drive of the replica node; flagging the dirty data as nondirty; broadcasting the information to associated nodes of the datacenter cluster; and writing back dirty data from respective solid-state drives of the associated nodes of the datacenter cluster to respective hard disk drives of the associated nodes. ### Claim 5 (depends on 4) The method of claim 4, further comprising: flagging data in the solid-state drives of the associated nodes with a nondirty flag; replacing the solid-state drive on the primary node; continuing to write incoming I/O requests to the solid-state drive of the primary node and the solid-state drive of the replica node; and writing new replica data from the associated nodes to the solid-state drive of the primary node. ### Claim 6 (depends on 1) The method of claim 1, wherein the detected failure comprises a cache device failure on the first replica node of the datacenter cluster, the method further comprising: disconnecting a primary node of the datacenter cluster; writing back dirty data from a solid-state drive of the primary node to a hard disk drive of the primary node; finding a remote second replica node of the datacenter cluster; storing the dirty data in a solid-state drive of the second replica node; flagging the dirty data in the solid-state drive of the second replica node as nondirty; and continuing to write incoming I/O requests to the solid-state drive of the primary node and the solid-state drive of the second replica node. ### Claim 7 (depends on 6) The method of claim 6, wherein finding a remote second replica node of the datacenter cluster comprises using a dynamic evaluation process. ### Claim 8 (depends on 1) The method of claim 1, wherein the detected failure comprises a communication failure between a primary node of the datacenter cluster and a first replica node of the datacenter cluster, the method further comprising: writing back dirty data from a solid-state drive of the primary node to a hard disk drive of the primary node; finding a second replica node to replace the first replica node; writing I/O requests to a solid-state drive of the primary node and a solid-state drive of the second replica node; and broadcasting to all nodes of the datacenter cluster a request to release all old replica data corresponding to the first replica node. ### Claim 9 (depends on 8) The method of claim 8, wherein finding the second replica node comprises using a dynamic evaluation process.