Warm Migration questions - Intel to AMD
-
So I just built some AMD powered small PCs to replace my big servers in my lab, and I'm getting things ready to move VMs from the old Intel to the new AMD (emphasis on lab). I thought the process was to warm migrate from one processor family to another, and maybe I'm doing something wrong.
XO from sources, XCP-ng 8.3 from a newly downloaded ISO and patched as of a couple hours ago.
I managed to get the pool master moved to an AMD server, and then picked on one of my throw away Windows Server VMs to try and migrate.
In XO I clicked on the VM, went to advanced, and clicked on Warm Migrate. This asked me which SR I wanted, delete the old VM, and start the new VM. I hit the go button and all it seems to be doing is moving the VM from SR to SR, which in this case is the same for both parts. Is this the expected result with XO from sources or did I do something wrong?
I was following this blog post https://xen-orchestra.com/blog/warm-migration-with-xen-orchestra/ which suggests I did the correct things, but the VM landed back on an Intel based server. I did a quick search of the documentation and didn't find what I'm looking for, but it was only a quick search and I may have missed it, google wasn't helping me today.
Should I just do a "standard" live migration to the AMD hosts? Or should I shut down the VM and use "start on XYZ host"?
-
What type of storage are you using? FYI, shared storage is only accessible by hosts within the same pool. You can't safely share the same SR with multiple pools.
From your description, it sounds like the warm migration was proceeding as expected. The VMs will crash if you live migrate them because of the difference in supported operations between the different CPUs.
-
Don't mix pools between Intel and AMD. Use warm migration to migrate from an Intel to an AMD pool.
-
I was wondering if that might be an issue. I will separate them back out and build an AMD pool and try again.
I may also have a network issue. The Intel hosts only have two connections, the AMD hosts have four connections. I'll detail that later after I try separate pools if I think it is still a problem.
-
As separate pool, this won't be a problem at all
-
Well... That worked just like you would think.
Set up a new NFS share on the new mini lab, set up a new pool made up of the three T740, selected my test machine and hit the warm migrate button to migrate to that new NFS share. After 22 minutes the VM popped up on the new system and the old one is waiting to be destroyed, just like planned. My only request would be to add the selection of the receiving host (if possible) this one migrated to the AMD master and I probably would have sent it to #2 or #3 since the master does more work than the other hosts.
Word or warning: Do not mix AMD and Intel on the same pool, and move the master function between them. I had to do a full shutdown to get control back so I could remove the AMD machines from the pool.
Also of note: When a host joins a pool, it will inherit the password from the master of the pool. All my AMD had a different password when I started this yesterday, now they share the same password as my old Intel master.
Went from this:
To This:
The first three on the bottom shelf are XCP-ng, the top three are vmware. The extra 2 are undecided, one might be a general Linux machine and probably XO from sources, the last might be a firewall since it has 6 network connections. Truenas Scale on the storage machine, 6 SATA spinning disks. All of this on 10 gigabit connections with a CRS326-24s+2q+ switch.
-
I have another question:
If I was going to move my production system from Intel Scalable Silver to Intel Xeon v2 to get my servers shut down for some construction, would warm migrate be the best way to move them?
We might be doing some physical construction in my server room, this would involve knocking out a concrete wall, and adding another concrete wall splitting the room in half. HVAC will also be shut down while this is happening. My choices are physically move my production servers, or rent/sell work my old lab servers. I have a room in another building where I'm going to be putting the servers that are really needed to keep things running.
What I'm thinking is that since there is a pretty big age gap in the processors, that a warm migration might be the best way forward to make sure that the network interfaces, and the processor architecture don't get in the way. Yes new shiny tool gets used for everything.
Also the Lab migrations are going really well, seeing some speed increase now that I have an XO on the new hosts and this XO has a 10gbps connection. Still only getting around 1-2gbps, but that's WAY better than the 400mbps I was getting before. Making me rethink how I have my production system running, and making me think about what I'm going to do with that fourth T740 in the picture (think it will be a standalone XOsources on 10gbps). I'm down to a Windows server of around 60GB disk was about 4 minutes, around 20 minutes the old way. Even my KASMweb server was around 4 minutes and it had around 500gb sized disk (not too much in use). Over all a big step up with small sized equipment. Even the Realtek gigabit NICs are working fine, can't say that with vmware!
-
Warm migration is probably the safest bet in all circumstances because there's no CPU compatibility issues
