KVM VM Replication
-
Basically, DRBD is way more advanced, and way faster, but has some fragility issues, too.
-
@fateknollogee said in KVM VM Replication:
What did you think?
How does it compare to Hyper-V replication?@Tim_G any updates ?
Sorry, Very busy day today.
It seems like a slight pain (but doable) to set up.
OVirt does it easy, so I'd go that route. Doing it that way at least looks better than Hyper-V, but Hyper-V is definitely easier than doing DRBD.
-
When all is said and done, I'd rather use KVM... if that answers your question.
-
http://yallalabs.com/linux/how-to-install-and-configure-drbd-cluster-on-rhel7-centos7/
http://www.learnitguide.net/2016/07/how-to-install-and-configure-drbd-on-linux.htmlNow this instruction is based on Proxmox. Its a Debian system so if you were do this on Fedora you will have to make several changes.
https://pve.proxmox.com/wiki/DRBD -
I'm not so focused on replication with KVM. It's not really needed. It's only hardware redundancy, and there is guaranteed data loss if you need to spin up a replica.
With Hyper-V, you can do replication of a VM every 30 seconds to 15 minutes. You could potentially lose 15 minutes of data.
If a server part went bad and the server died, I'd rather let the VMs on it be down for a few minutes to replace the part, and then bring everything back up again... rather than failover every VM to the replica and lose all that data.
With KVM, it's like Scott said... you can do replication but it's meant to be a network RAID1. That's HA. It's better to go that route instead. With Linux, it's free (if you exclude hardware costs because you may re-use something you already have).
-
Any improvements on the KVM + Replication front?
-
@FATeknollogee said in KVM VM Replication:
Any improvements on the KVM + Replication front?
Improvements? What would you want improved? Where do you perceive it not having met any potential need all along?
Where this thread went was... KVM was so good at replication that it didn't need the fall back methods popular on Hyper-V that exist to make up for the lack of HA full sync replication that KVM and Xen had.
-
@scottalanmiller You aren't talking about the DRBD in HA-Lizard?
-
Scratch my last post, Lizard is XS only.
-
@FATeknollogee said in KVM VM Replication:
@scottalanmiller You aren't talking about the DRBD in HA-Lizard?
DRBD is part of the Linux kernel, so KVM has it natively. Xen typically gets it, but Xen isn't tied to Linux, so not always. But KVM and DRBD are always linked. So KVM always has DRBD no matter how it is installed unless you compile it out on purpose, and no one does that.
-
DRBD has way more than that, though. All normal KVM installs have options like Gluster and CEPH, too. And that's just the included replication options.
If you want to go commercial, there are options like Starwinds VSAN.
So KVM replication is essentially the market leader with the most and best integrated features, and the most and best commercially available features.
Basically anything you want, KVM does as well or better than everyone else.
-
Who amongst us is doing is currently doing KVM replication?
-
@FATeknollogee said in KVM VM Replication:
Who amongst us is doing is currently doing KVM replication?
We do it with SCRIBE on our Scale HC3 system.
-
@FATeknollogee said in KVM VM Replication:
KVM replication
I think calling it this adds to the confusion. Hyper-V includes a product called Hyper-V Replication. This makes it seem like it is both a part of the hypervisor (it is not) and something we should be looking for as a feature (it is not.)
No other platform has a thing called "platform replication". It's just not an appropriate concept. And Hyper-V only had it to overcome storage layer shortcomings that were pretty dramatic compared to the competition. And now in Hyper-V, you basically never use it because it's not a good idea, there are free add ons that make it essentially (but not absolutely totally) useless.
In KVM, Xen, or ESXi we should simply be asking about replication strategies. How do you want to replicate, and why? How do you do that with your specific KVM? KVM is just the hypervisor and nothing else, but Hyper-V is the name for the entire ecosystem around a hypervisor. So asking about KVM as if it were similar doesn't make sense.
If what you want to ask about is DRBD, ask about DRBD, not KVM. If you want to know about Gluster, ask about that. Asking about a storage feature concept, but only in the context of a non-storage system, will result in confusing and very limited answers.
-
@scottalanmiller said in KVM VM Replication:
No other platform has a thing called "platform replication".
vSphere Replication is most certainly a product/feature, and other ecosystem replication solutions out there like Recovery Point, RP4VM, Veeam Backup and Replication and Zerto. Couple things...
-
DRDB is a kernel driver. There's positives (It's low overhead) but also negatives vs stuff like VAIO that loopback to userspace (if it borks it takes down the host). For async you don't want WAN backups or other issues to cause problems, and how you deal with resume from the replication being borked without having something like CBT or WAN acceleration can change hugely how the stuff is operationally managed.
-
Compression/Deduplication/WAN efficiency. In theory I could run DRDB through a WAN optimizer, but this is a standard function in a number of replication products. If your on metered connections this can get expessive. (vSphere Replication uses LZFast for compression, Veeam lets you tune what compression you want etc). Honestly don't remember what compression Hyper-V uses other than that they do it.
-
Encryption. A lot of other async replication products/features support TLS encryption at the transport layer for this connection. In theory you could IPSEC DRDB.
-
Does the product support GFS retention of the async replica's? Can I keep snapshots on the other side and is merging them out to age out old ones a zero overhead operation.
-
Application and guest quiescence etc tied to recovery points. What does management for this look like?
-
Can I target 1 cluster to another cluster without having to manually pair hosts and volumes? (IE can I just target a resource pool etc on the other side, or am I manually deciding target hosts and volumes every time).
Some other nice to have's is something that manages orchestration of testing, failover, failback with full runbooks, replication of network firewall rules, handling BGP failover, network overlay stuff etc.
Saying "It's cool DRDB can do async replication" may be TECHNICALLY true, but there's a rabbit whole of features people tend to look for with async replication
-
-
@brianlittlejohn said in KVM VM Replication:
I may be asking this prematurely, I'm about to start playing around with KVM. Is there a replication feature similar to HyperV replication baked into KVM natively?
Many times we soo those questions, Perhaps if we can PIN KVM as topic as well as (pbx) freepbx some how
-
@scottalanmiller tags please
-
-
@Emad-R said in KVM VM Replication:
@brianlittlejohn said in KVM VM Replication:
I may be asking this prematurely, I'm about to start playing around with KVM. Is there a replication feature similar to HyperV replication baked into KVM natively?
Many times we soo those questions, Perhaps if we can PIN KVM as topic as well as (pbx) freepbx some how
Not realistically. If we started pinning random topics, the topic list would turn into a handful of tagged topics and it would be very diffult to use the forum. This is such an obscure topic, we'd likely pin a hundred or more topics before this, imagine how hard it would be to use a forum with a few hundred pinned topics that you always saw. You'd never see anything else, nor could you find what you needed in the pinned list.
Just using the search is best.
-
