Server 2019: Ppor Storage Spaces Direct (S2D) Performance
Hello, I am trying to setup a two-node cluster for S2D for hyperconverged infrastructure. I am underwhelmed by the performance when compared with storage performance when the drives are not part of an S2D cluster and some older hardware we have on hand. I want to ensure I am not missing anything. Environment Node 1: System: Dell PowerEdge R730 CPU: (2) Intel Xeon E5-2660 RAM: 256 GB Node 2: System: Dell PowerEdge R630 CPU: (2) Intel Xeon E5-2698 v4 RAM: 256GB OS: Windows Server 2019 Data Center Network: 10G Ubiquiti USW-Pro-Aggregation switch Each node has (1) Intel Ethernet Converged Network Adapter X710 set as the management NIC Each node has (4) QLogic 10GE QL41164HMCU CNA ports connected to the Ubiquiti switch The 4 NICs are put into a Switch Embedded Teaming (SET) switch RDMA is enabled and set to iWarp. Have verified RDMA connections Storage: Each node contains the following drives: (2) Intel Optane DC 4800X SSDPED1K375GA 375GB SSD PCIe (4) Samsung MZILS3T8HMLH0D3 3.49TB SSD SAS (2) Toshiba KPM5XRUG3T84 3.49TB SSD SAS (5) Toshiba AL15SEB18EP 1.64TB HDD SAS I have been using DISKSPD and the output is below for the cluster using the entire storage pool. I would have expected some higher MiB/s and I/O per s. If I change the config and use only SSDs like even just one of the Samsung drives I get about 900MB/s and 15K+ I/O per s. The performance is about the same as isolating one of the Toshiba HDDs and running off there which I tried as well. I would have expected it to be much faster. * This specific example is where I am trying to test write performance but I am similarly underwhelmed by read performance as well. diskspd -b60K -d60 -h -L -o32 -t1 -s -w100 -c1G C:\ClusterStorage\HV1\LogIO.dat Total IO thread | bytes | I/Os | MiB/s | I/O per s | AvgLat | LatStdDev | file ----------------------------------------------------------------------------------------------------- 0 | 21557882880 | 350877 | 342.66 | 5847.99 | 5.471 | 1.226 | C:\..\LogIO.dat (1024MiB) ----------------------------------------------------------------------------------------------------- total: 21557882880 | 350877 | 342.66 | 5847.99 | 5.471 | 1.226 Read IO thread | bytes | I/Os | MiB/s | I/O per s | AvgLat | LatStdDev | file ----------------------------------------------------------------------------------------------------- 0 | 0 | 0 | 0.00 | 0.00 | 0.000 | N/A | C:\..\LogIO.dat (1024MiB) ----------------------------------------------------------------------------------------------------- total: 0 | 0 | 0.00 | 0.00 | 0.000 | N/A Write IO thread | bytes | I/Os | MiB/s | I/O per s | AvgLat | LatStdDev | file ----------------------------------------------------------------------------------------------------- 0 | 21557882880 | 350877 | 342.66 | 5847.99 | 5.471 | 1.226 | C:\..\LogIO.dat (1024MiB) ----------------------------------------------------------------------------------------------------- total: 21557882880 | 350877 | 342.66 | 5847.99 | 5.471 | 1.226 I would appreciate any help. Thanks!.
Cannot create separate storage pool on Windows 2019 Failover cluster.
Hello I am trying to get rid of S2D on our failover cluster and came across the inability to create separate storage pools (without using S2D) on Windows 2016/2019. Every time you set up a Failover Cluster, even without adding eligible disks to storage, it still sets up the Clustered Windows Storage subsystem and assigns all available disks to it. So, you have all the disks, but if you try to create a storage pool, it will fail. Tried to create all storage pools before forming the cluster - to no avail, the pools were created, but any attempt to create a virtual disk instantly fails. I tried to use Set-ClusterS2DDisk to stop the storage subsystem from using those disks - to no avail. Tried to unregister the Clustered Windows Storage subsystem - it's not possible. I have absolutely no idea what to do now. So, is there a way to set up a failover cluster without using S2D?
Feature request: s2d manual cache binding
Hello! Please make available tools for manual binding of cache drives to data drives. First problem: In configurations with multiple storage pools cache drives are bound to all physicall disks, not just to disks in pool. Use cases: binding problems. in that cases specialists are needed to dance with tambourine to workaround that.the're evicting nodes/retiring drives and so on. that is really PITA. Second problem: set-clusterstoragespacesdirct -cachestate enabled resets the states of journal drives. Use case: you want to dedicate some drives to hot-spare, after enabling cachestate, all drives became journal and bound, that is really not what you want if you're planning to use hotspare drives, especially if you're using M.2 SSDs
