Hi! for a small to medium high availability cluster based on simple server/standard hardware (SATA drives), we build an DRBD8 Cluster with 2 Nodes.
2 x Hardware (2 x mdm raid 5) 2 x DRBD /dev/drbd0 running nearly fine File system on DRBD Resource today ocfs2. Kernel on Host: 4.20.6, 8 CPU Scheduler on Host: mp-deadline Network for DRBD dedicated, ptp, 10GBe. virt hosts are managed by libvirt/kvm/qemu: Guests: Debian Jessie, Stretch, Buster (qcow2, scheduler noop/none, virtio-net, virtio storage, file system btrfs noatime,nodiratime) And on demand: Win7, not permanent online This solutions works nearly fine, but: 1. The available storage bandwidth in the guests is max 30 M Byte/sec, looks a little to slow. 2. On heavy I/O like BTRFS operations in one guest (scrub, balance, de-fragment), the host gets a huge load problem, the host load runs against 40, 50, 60, 80 ..., even iothreads are set to 4. The guest latency is arisen a lot, nfs clients to the guest are spamming "nfs server not responding" ... After a few minutes the load impact solves by itself... 3. sometimes the guest looses network connectivity, even during high load situations on DRBD, like read of drbd8/ocfs2 and write backup to external disks. 4. On OCFS2 the read operations are breaking down to 10, 20 M Byte/sec during rsync backups of the DRBD8/OCFS2 Resource. About 99% of production time, all is really fine, we are worrying about the 1% in case of running backups and the side effects of the load. Any ideas? Has anybody seen those effects in practice? mit freundlichen Grüßen Jürgen Sauer -- Jürgen Sauer - automatiX GmbH, +49-4209-4699, juergen.sa...@automatix.de Geschäftsführer: Jürgen Sauer, Gerichtstand: Amtsgericht Walsrode • HRB 120986 Ust-Id: DE191468481 • St.Nr.: 36/211/08000 GPG Public Key zur Signaturprüfung: http://www.automatix.de/juergen_sauer_publickey.gpg
<<attachment: juergen_sauer.vcf>>
_______________________________________________ drbd-user mailing list drbd-user@lists.linbit.com http://lists.linbit.com/mailman/listinfo/drbd-user