[ovirt-users] NFS sync vs async
Hello there, At first I thought I had a performance problem with virtio-scsi on Windows, but after thorough experimentation, I finally found that my performance problem was related to the way I share my storage using NFS. Using the settings suggested on the oVirt website for the /etc/exports file, I implemented the following line: /storage *(rw,sync,no_subtree_check,all_squash,anonuid=36,anongid=36) The underlying filesystem is ext4. In the end, whatever the VM I am running through this NFS export, I get extremely poor write performance, like sub-100 IOPS (my disks usually can do 800-1k). Under the hood, iotop shows that my host IO is all taken up by jbd2, and if I understand correctly, it is the ext4 logging process. I have read that using the "async" option in my NFS export is unsafe, like if my host crashes during a write operation, it could corrupt my VM Disks. What is the best combination of filesystem / settings if I want to go with NFS sync? Is someone getting good performance with the same options as me? if so, why do I get such abysmal IOPS numbers? Thanks! J-F Courteau ___ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
[ovirt-users] Windows 2012R2 virtio-scsi abysmal disk write performance
Hello there, I just installed oVirt on brend new machines. The engine on a Virtualbox VM in my current infrastructure, and a single dedicated CentOS host attached to the engine. Here are my host specs (srvhc02): CPU: Intel(R) Core(TM) i7-4790 CPU @ 3.60GHz (4C 8T) Mem: 32GB DDR3 Disks: - 2 x WD Black 2TB (RAID1) where the CentOS is installed - 4 x WD Gold 4TB (RAID10) dedicated for a storage domain (/dev/md0, ext4, mounted in /storage and exported with NFS with default settings) NIC: Intel 4 ports Gigabit. One port for VMs, one port for everything else. 2 unused ports. OS: Freshly installed CentOS 7 minimal latest (yum updated) oVirt deployed from the Engine using root account with a password Here are my Engine specs (srvvmmgmt): Virtualbox guest VM in another physical computer just for test purposes. CPU: 2vCPU Mem: 8GB Disk: 50GB VDI on a RAID10 array NIC: Virtual Intel Pro 1000 (Bridged over a physical Intel 4 ports Gigabit) OS: Freshly installed CentOS 7 minimal latest (yum updated) oVirt Engine 4.2 deployed from the repo, yum updated yesterday Storage domain: NFS - srvhc02:/storage ISO domain: NFS - srvhc02:/iso Physical network is Gigabit on Cisco switches, no VLAN tagging. My (barely usable) Windows 2012R2 guest (srvweb03) CPU: 2vCPU Mem: 8GB Disk1: 100GB on the storage domain Disk 2: 200GB on the storage domain (this is the one I was changing the controller for testing purposes) NIC: virtio bridged over the VM port I have tried every possible combination of drive controller (virtio-scsi, virtio, IDE), virtio drivers (stable and latest from the Fedora website, others found in sometimes obsucre places), and the disk write performances are simply unacceptable. Of course I had a reboot between each NIC driver update and driver change. I have compared with a Virtualbox installed on the same host after I cleaned up oVirt, and the host is absolutely not problematic. Here is the compare: oVirt Virtualbox --- SEQUENTIAL READ 3000MB/s 406MB/s (this is ridiculously high, oVirt probably reading from RAM) SEQUENTIAL WRITE 31MB/s164MB/s 4k RANDOM READ400MB/s 7.18MB/s(this is ridiculously high, oVirt probably reading from RAM) 4k RANDOM WRITE 0.3MB/s(80IOPS) 3MB/s (747IOPS) These performances are using the virtio-scsi in oVirt (any version, none gave me better performance), and the SATA driver in Virtualbox. Needless to say that this is frustrating. I have tried with oVirt virtio-blk (virtio in the interface) driver, it gave me a 10-15% improvement, but never up to the level I could expect from an enterprise grade solution. I also tried IDE. I could not even write at 10kbps on it. I had approx 10IOPM (yes, IO per minute!) with it. Is there something I am missing? -- https://nexcess.ca JEAN-FRANÇOIS COURTEAU _Président, Directeur Général_ C: jean-francois.court...@nexcess.ca T: +1 (418) 558-5169 - ___ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
[ovirt-users] Windows 2012R2 virtio-scsi abysmal disk write performance
Hello there, I just installed oVirt on brend new machines. The engine on a Virtualbox VM in my current infrastructure, and a single dedicated CentOS host attached to the engine. I am getting extremely poor write performance in my oVirt Windows 2012R2 VM, whatever the virtio, virtio-scsi device I use, and whatever the virtio Windows driver version. Here are my host specs (srvhc02): CPU: Intel(R) Core(TM) i7-4790 CPU @ 3.60GHz (4C 8T) Mem: 32GB DDR3 Disks: - 2 x WD Black 2TB (RAID1) where the CentOS is installed - 4 x WD Gold 4TB (RAID10) dedicated for a storage domain (/dev/md0, ext4, mounted in /storage and exported with NFS with default settings) NIC: Intel 4 ports Gigabit. One port for VMs, one port for everything else. 2 unused ports. OS: Freshly installed CentOS 7 minimal latest (yum updated) oVirt deployed from the Engine using root account with a password Here are my Engine specs (srvvmmgmt): Virtualbox guest VM in another physical computer just for test purposes. CPU: 2vCPU Mem: 8GB Disk: 50GB VDI on a RAID10 array NIC: Virtual Intel Pro 1000 (Bridged over a physical Intel 4 ports Gigabit) OS: Freshly installed CentOS 7 minimal latest (yum updated) oVirt Engine 4.2 deployed from the repo, yum updated yesterday Storage domain: NFS - srvhc02:/storage ISO domain: NFS - srvhc02:/iso Physical network is Gigabit on Cisco switches, no VLAN tagging. My (barely usable) Windows 2012R2 guest (srvweb03) CPU: 2vCPU Mem: 8GB Disk1: 100GB on the storage domain Disk 2: 200GB on the storage domain (this is the one I was changing the controller for testing purposes) NIC: virtio bridged over the VM port I have tried every possible combination of drive controller (virtio-scsi, virtio, IDE), virtio drivers (stable and latest from the Fedora website, others found in sometimes obsucre places), and the disk write performances are simply unacceptable. Of course I had a reboot between each NIC driver update and driver change. I have compared with a Virtualbox installed on the same host after I cleaned up oVirt, and the host is absolutely not problematic. Here is the compare: oVirtVirtualbox --- SEQUENTIAL READ *3000MB/s 406MB/s SEQUENTIAL WRITE 31MB/s164MB/s 4k RANDOM READ *400MB/s 7.18MB/s 4k RANDOM WRITE 0.3MB/s(80IOPS) 3MB/s (747IOPS) *this is ridiculously high, oVirt probably reading from RAM These performances are using the virtio-scsi in oVirt (any version, none gave me better performance), and the SATA driver in Virtualbox. Needless to say that this is frustrating. I have tried with oVirt virtio-blk (virtio in the interface) driver, it gave me a 10-15% improvement, but never up to the level I could expect from an enterprise grade solution. I also tried IDE. I could not even write at 10kbps on it. I had approx 10IOPM (yes, IO per minute!) with it. Is there something I am missing? Thanks! ___ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
