Search results

Well, I had in another DC (in Germany but not Hetzner, though they're pretty good as well) a preconfigured SWRAID setup. I requested a rescue boot (they do it via PXE IIRC), deconfigured their swraid using LSI command line tools and installed Debian via debootstrap, booted and then created a zfs...

Yes, sory for being imprecise, 1 zvol per VM disk. zvol has the supposed advantage of being a "simple" block device with fewer layers of complexity and thus being faster for this kind of storage, but it's better to test in your own systems first. The PVE defaults are pretty good though, I rarely...

Both is supported. You can either use raw file images in directory storage in zfs or zvol block device-based images. The default config uses the latter (zvols for each vm).

I usually get great performance using normal sas/sata disks with slog and l2arc on ssd and using zvols. Qcow2 or raw in directory performs worse. I get better performance using virtio vs. virtio-scsi in Windows. This migt vary somewhat per system but I suggest trying zvols, too, for vm images.

Sorry for reviving this old thread but I'm wondering if you found a solution to the performance problem on this particular Dell server.

Well, until then there's a simple method for scripting using pvesh. It works fine and will suffice for my current needs. It does't work with pve-zsync so I'll need to use my own zfs backup wrapper. That should be addressed when pve-zsync natively supports this feature someday.

Great! Can't wait to test it.

That's nice to hear! Until then, I'd like to script it for my needs. Can you recommend a way to acheive that by using the PVE API, command line or such? I've never used the API and related facilities yet, but I'd just need a small initial guidance on where to start, if possible.

The pve-zsync tool is a very useful addition. If I'm planning to use it for regular VM snapshots, can it or does it issue an FS freeze when using the qemu guest agent? In other words, will I have such a consistent backup like when using vzdump with the agent?

Or use an USB over IP solution, like this. It's supported on Windows and Linux on multiple architectures. You might be able to install the server component on the HN and the client in the VM. Just an idea.

If you're using a separate SATA card for your setup, you can blacklist the driver module for it. Eg. plut "blacklist mpt2sas" (use the module for your card) in /etc/modprobe.d/pci-redirect.conf and reboot.

I finally got around to test this series, I gotta say, lots of work, congrats! I for one like the new columnised design, for me it's much quicker to navigate and more straightforward, a lot less mouse movements are needed. I was "frightened" first seeing the heaps of negative reactions, but it...

I'm aware, that's why I wrote the next part of the sentence: "which might not be a viable approach with LXC". I don't expect it to be implemented, but it's very convenient in OVZ. As things stand, not even OVZ 7 supports it, as you hinted, too. Ah well. BTW, it's nice to hear you already work...

What I miss exactly: represent the number of CPUs and memory size consistently across procfs and kernel api, at the least, without CPU pinning or other quirks. Something like OVZ has provided since forever. Simple settable quotas would be nice but I understand OVZ used simfs for that which might...

Yes, I know, but it doesn't fully represent resource settings inside the CT - that's my basic problem with LXC and its tools. Otherwise it's pretty stable and works as intended, but until resource limit presentation is not fixed, I'm stuck with OVZ with most installations. Well, I haven't even...

@dietmar: you talk like it's an easy task for everyone to become a kernel/container developer :) Well, it's not ant I'm not one. I understand Proxmox won't have support for it, but I thought I give a shot at asking. What I miss from LCX is more functions for isolation, like showing only the...

I see. With a single-vdev raidz2 speed would be slow for writing generally since it would be as slow as a single component disk. ZFS really shines with many vdevs, so eg. with multiple raidz2 vdevs striped, and possibly cache device(s) added. I think it would be a fair comparison then. Otherwise...

Well, for many it might not be exactly news but the new release of OpenVZ/VIrtuozzo 7 is out as some kind of obscure standalone distro. Is anyone using it, or are there any plans to integrate it in Proxmox at some point in time? I know it's on a different road now with LCX but many important...

What do you need speed for? Can we have a glimplse of your planned setup? It won't be a double in speed difference but IOW will be much higher. Is that even important for you?

But ZFS is not just about speed, it's a lot of added functionality you will never have with any HWRAID card. Defining a single criteria like speed makes no sense in my eyes. After judging that HWRAID is faster on your own system by 5% in synthetic tests, does that clearly indicate you'll benefit...