Технологии работы с дисковыми хранилищами и файловыми системами Windows Server 2016 (Часть 2)
0 likes1,870 views
The document discusses the features and functionalities of Storage Replica, including its capabilities for synchronous and asynchronous replication over SMB 3.1.1. It highlights various use cases, management tools, performance metrics, and distinctions from other replication technologies. Additionally, the document emphasizes the importance of system requirements, log management, and optimizations for effective data replication across clusters.
Технологии работы с дисковыми хранилищами и файловыми системами Windows Server 2016 (Часть 2)
- 7. Replication Block-level, volume-based Synchronous & asynchronous SMB 3.1.1 transport Write ordering Seeded and thin provisioned Flexibility Any Windows data volume Any fixed disk storage Any storage fabric Shared cluster storage S2D cluster storage Management Failover Cluster Manager Windows PowerShell Azure Site Recovery End to end MS storage stack
- 8. SR and Storage Spaces Direct close the loop on the MS storage stack
- 11. Single cluster Automatic failover Synchronous & Asynchronous Asymmetric storage Two sites, two sets of shared storage Cluster storage: CSV or role-assigned PDR Manage with Cluster Manager Or Windows PowerShell Increase cluster DR capabilities Hyper-V and General Use File Server are the main use cases New York New Jersey SR over SMB The Hudson River
- 12. Two separate clusters Manual or orchestrated failover Synchronous or asynchronous S2D and shared disk supported Manage with PowerShell & ASR Los Angeles Las Vegas SR over SMB
- 13. Two separate servers Manual failover Synchronous or asynchronous Manage with PowerShell …only? Building 5 Building 9 SR over SMB
- 14. Remote Server Management Tool support for SR Server to server to start Cluster to cluster to follow Fully integrated into RSMT Coming at or slightly after TP5
- 18. This is not DFSR This is not DFSR! Replicating storage blocks underneath the CSVFS, NTFS, or ReFS volume Don’t care if files are in use Write IOs are all that matter to Storage Replica
- 19. V o l u m e
- 20. Applications (local or remote) Source Server Node (SR) Data Log 1 t 2 Destination Server Node (SR) Data Log t1 3 2 5 4
- 21. Applications (local or remote) Source Server Node (SR) Data Log 1 t 2 Destination Server Node (SR) Data Log t1 5 4 3 6
- 22. 1. Replicating normally 2. Destination offline, the source log contains all IO: A. Replay from log B. Destination comes back online (faster) 3. Destination offline, the source log wrapped IO A. Replay from bitmap B. Destination comes back online (slower) Source Server Node (SR) Data Log Destination Server Node (SR) Data Log OI Replay Log DeltaReplay Bitmap Delta The lesson: the bigger the log, the faster the recovery. And we never block IO
- 23. Using the scalability and perf built into SMB SMB Multichannel SMB Direct (RDMA) We are currently testing Mellanox InfiniBand MetroX and Chelsio iWarp configs We have several 10km, 25km and 40km networks Roundtrip latencies: ~200usec at 80km Encryption and signing 3.1.1 encryption performance improvements (awesome parallelization)
- 25. Datacenter Edition (Full, Core, and Nano) Active Directory ≥4GB RAM, ≥2 Cores Disks Firewall ports
- 26. Test-SRTopology cmdlet Checks requirements and recommendations for bandwidth, log sizes, IOPS, firewall ports, etc… Tells you how long initial sync should take Gives you a tidy HTML report with recommendations Customers love it When they find it
- 29. Destination volume is always dismounted One to one There is no “test failover” Hmmm Mixed cluster and standalone blocked Currently All of above planned for post-RTM Don’t let your customer write checks we can’t cash
- 30. Async crash consistency versus application consistency We guarantee a mountable volume Your app must guarantee a usable file Volume Shadow Copy Snapshots They replicate too Accept that async means possible data loss How much money is your data worth? Or your job?
- 31. Hyper-V Replica Yes NA Yes (VMs hosting file servers) No Yes (VMs hosting SQL) Yes (VMs hosting apps & servers) VM SYSVOL File Server MS Exchange SQL Server Other Applications Storage Replica Yes No Yes (bare metal or VMs) No Testing (bare metal or VMs) Yes (bare metal or VMs) SQL Server AlwaysOn FCI NA NA NA NA Yes NA MS Exchange DAG NA NA NA Yes NA NA DFS Replication No Yes Yes No No No FRS Never Upgrade to DFSR, soon Don’t do it Are you crazy? What? No! Hahaha, you are funny SQL Server AlwaysOn AG NA NA NA NA Yes NA Workload Replicator
- 32. Hyper-V Replica advantages and characteristics Use SR when you need:
- 33. In a VM In Azure Will we backport SR to Windows Server 2012 R2? Please?
- 34. Storage Replica is not “shared nothing” clustering Storage Replica is not a backup Storage Replica is not DFSR Storage Replica is not a general branch office solution
- 42. http://blogs.technet.com/b/filecab/archive/2015/07/09/storage-replica-versus-robocopy-fight.aspx Technique Data to copy Time to complete Robocopy /s /copyall /zb /mt:128 95GiB, 100M files 8 hours, 0 minutes and 12 seconds Robocopy /copyall /zb 1.45TiB, 1 file 2 hours, 48 minutes, 26 seconds Robocopy /copyall /j /zb 1.45TiB, 1 file 2 hours, 38 minutes, 8 seconds Robocopy /s /copyall /zb /mt:128 1.45TiB, 1.11M files 25 minutes, 0 seconds Robocopy /copyall /j 1.45TiB, 1 file 15 minutes, 23 seconds Storage Replica 1.45TiB, the volume 14 minutes, 27 seconds
- 43. 4 Intel(R) Xeon(R) Processors (16 cores) o Intel® Xeon® Processor E5-2623 v3 (10M Cache, 3.00 GHz) Installed Physical Memory (RAM) 128 GB (24) x 2.5" SAS/SATA drive bay, Hot-swap with 12Gb/s backplane (4) x Redundant cooling fans (2) Risers x8 slots low profile PCIe Gen3 (2 available per node) (2) Intel(R) 10Gb Ethernet SFP+ connections on-board (per Node) (1) 1200W (1+1) Redundant Power Supply Remote Management Module, Integrated BMC w/IPMI 2.0 NIC port (4) External SAS Ports (per Node) (1) DataON 2U "Standard" RACK mount Rail set Internal OS Drives 4: Sandisk x300 128GB 2.5 SATA III MLC SSD - 2 per Node 12G SAS HBA for Data Drives - Avago(R) LSI Storage 9300-8i PCI-e 3 8-port internal 12Gp/s SAS HBA. Integrated mezzanine card 12G SAS Cabling for Expansion - SFF-8644 to SFF-8644 Ext. 2M 12Gb/s SAS Cable for expansion 12 1800GB HGST SAS 2.5" 10k RPM - HDD Part # Ultrastar C10K1800 21.6TB Raw per appliance 12 400GB HGST 12Gb Ultrastar SSD1600MM SAS 2.5" - SSD Part # HUSMM1640ASS04 4.8TB Raw per appliance Network Card(s): 7 NIC(s) Installed. o Intel(R) 82599 10 Gigabit Dual Port Network Connection o Mellanox ConnectX-3 Pro Ethernet Adapter o Chelsio Network Adapter o Chelsio Network Adapter o Intel(R) 82599 10 Gigabit Dual Port Network Connection o Mellanox ConnectX-3 IPoIB Adapter Mellanox ConnectX-3 IPoIB Adapter
- 44. Metric 8 KB I/O 16 KB I/O 32 KB I/O 64 KB I/O Percentage Write Latency Overhead due to SR -56% -45% -65% -66% Percentage Read Latency Overhead due to SR - 100% Read 7% 5% 2% 23% Average Write Latency to Source Log disk in milliseconds 1.24 1.53 2.78 3.7 Average Write Latency to Destination Log disk in milliseconds 1.17 1.35 2.81 3.31 I/O Size (KB) Read Latency - 100% Read (ms) Write Latency – 80% Write (ms) Without SR With SR Without SR With SR 8 5.37 5.77 41.98 18.31 16 6.08 6.37 42.79 23.68 32 4.8 4.92 52.16 18.12 64 5.52 6.77 68.45 23.49
- 47. Configure SR without being an administrator Pair two clusters
- 48. Replicate thin-provisioned Storage Spaces volumes Support for 3rd parties possible but requires validation
- 49. Pick your networks Limit by machine or by partnership Also limit bandwidth Set-SmbBandwidthLimit -Category StorageReplication -BytesPerSecond <uint64>
- 50. Change partition sizes mid-replication No need to recreate replication, not blocked while replicating ReFS is grow-only
- 52. Lots of issues in TP4 All working in TP5 Probably Please test and file bugs, Nano is very dynamic code
- 53. Hooked into new Cluster health system RPO dial for async Not a bandwidth limiter More of an early warning system Under-provisioned log warnings (maybe not TP5)