I installed in a an HP Gen 10 with 512Gb of RAMM and dual Gold CPU a LQD3000 card (4 Commercial Samsung Disk of 1.92Tb). I then made a RAID10 (using the 4x1.92Tb) Giving me around 3.70TB). When I do Crystal Mark speed test in a Windows VM (and the Virtuo Drivers have been installed) I get really slow random read/write speed. Is there something I am missing for configuration or that I should change in the machines configuration?
Random 4k read and write slow
- Thread starter JSChasle
- Start date
-
- Tags
- drive speed raid 10
Really thought these card (at their price point) Would have higher I/O speed. I get about the same speed using Red NAS NVME at a fraction of the price...
What model are the disks? It is a well-known fact that for both virtualization workloads and ZFS, commercial disks can get very slow really quick.
The drive inside de LQM3000 are Samsung MZ1LB1T9HALS-00007 https://semiconductor.samsung.com/ssd/enterprise-ssd/pm1733-pm1735/mzwlj1t9hbjr-00007/ and card is an https://assets-global.website-files...3b5e451187ae_Liqid-Element-LQD3000-050420.pdf
The Liqid Element PCIe Add-In-Card (AIC) features high dense capacity and extreme performance for mission critical and performance-demanding workloads. It is an ultra-thin, standard form factor half-height half-length (HHHL) card that works seamlessly with systems that have existing PCIe slots. This makes the Element AIC ideal for deployment in data center and enterprise applications.The Element AIC offers a Gen 3.0 x8 PCIe interface, which enables high-throughput and low-latency transactions. It utilizes the latest NVMe protocol in order to deliver increased performance and efficiency from a single device. The AIC outperforms legacy architectures by delivering 1.25 M IOPS of random performance, over 7 GB/s of throughput and ultra-low transactional latency of 20 μs
please try test with:
zfs set atime=off relatime=off compression=off sync=disabled <yourpool>
Add to /etc/kernel/cmdline (or /etc/default/grub):
......... pcie_aspm=off intel_idle.max_cstate=0 intel_pstate=disable processor.max_cstate=1 mitigations=off
#update-grub
#update-initramfs -u -k all
#reboot
zfs set atime=off relatime=off compression=off sync=disabled <yourpool>
Add to /etc/kernel/cmdline (or /etc/default/grub):
......... pcie_aspm=off intel_idle.max_cstate=0 intel_pstate=disable processor.max_cstate=1 mitigations=off
#update-grub
#update-initramfs -u -k all
#reboot
is just the card which just passthrough the real ssd which are fast only within their unsecure cache which ZFS can't fully use. (sorry for my poor english wording).
ssd like Samsung PM9A3 is ready for zfs.
Hi will try it in off hours and after a backup (since 2 critical machines are being used during the day). Thanks for the input
any luck after applying the ZFS tweaks/driver changes? Did the random I/O improve or is it still slow?