BRRABill's Field Report With XenServer
-
@olivier said in BRRABill's Field Report With XenServer:
@BRRABill using a guest agent will be faster than a "copy" on hypervisor level (for the record, it uses compression and read the whole disk!)
Remember than a "copy" is a high level feature provided by XO, using streaming export from one server to VM import on the other side. With GZIP compression on XenServer. You bet it's slower than working on guest level!
If you want to compare backup speed, using continuous delta backup could be interesting: after the initial backup (which is not compressed), XO will only export delta (the diff written between the new backup and the previous one). And forever, because we (XO) will merge block of the oldest delta in the full (initial) backup. So no more full export after the first one.
So it's more comparable to your sector-level backup program (it uses the diff of the VHD format). The main difference is you don't need to install any program in your VM to have it working
Does export from XC definitely use compression? That is the one thing I think we were still debating in the other thread.
-
I'm not an expert in XenCenter, AFAIK there is a checkbox to compress or not when exporting.
-
@olivier said in BRRABill's Field Report With XenServer:
@BRRABill using a guest agent will be faster than a "copy" on hypervisor level (for the record, it uses compression and read the whole disk!)
That was something else I figured had to be happening, that is reads the entire disk.
Which, if you have a lot of empty space for whatever reason, adds a lot of extra time.
There is an instance it would be nice to be able to shrink the VD, then do the copy/export. Maybe a pipe dream!
-
IIRC, XenServer skip the 0's without compression (I'm not sure but I remember some stuff about the pipe needed to pass it to GZIP which need to handle the read differently than exporting directly). Anyway, continuous delta backup won't suffer of this because exporting the diff only.
-
@olivier said in BRRABill's Field Report With XenServer:
IIRC, XenServer skip the 0's without compression (I'm not sure but I remember some stuff about the pipe needed to pass it to GZIP which need to handle the read differently than exporting directly). Anyway, continuous delta backup won't suffer of this because exporting the diff only.
The reason I chose XC export instead of XO backup was because in my test VM (a Server 2012 core install) it took around 5 minutes to export, and 12 minutes to backup. I figured the backup would more than double the time I needed to accomplish my task.
But in the other thread (on performance) you said that backup should be quicker.
Any thought on why I saw opposite results? Was the sample too small? I'd be glad to do another test.
-
Again, I can't guess which kind of backup did you used, on which kind of storage, on which kind of network etc...
edit: it's like you told me, "my car is slow", without telling which car, on which road and in which situation
-
@olivier said in BRRABill's Field Report With XenServer:
Again, I can't guess which kind of backup did you used, on which kind of storage, on which kind of network etc...
edit: it's like you told me, "my car is slow", without telling which car, on which road and in which situation
WHY DO I KEEP DOING THAT?
I set up a one-time backup of the VM through XO.
The backup went to a Windows share. The same share I used for the export location.
-
But what kind of backup? Normal backup or Continuous Delta backup? If normal backup, did you deactivate the compression or not?
Also, exporting in your Windows share from XO as a backup could produce different result than using XS from your computer (not the same network path).
That's a lof of unknown
-
@olivier said in BRRABill's Field Report With XenServer:
But what kind of backup? Normal backup or Continuous Delta backup? If normal backup, did you deactivate the compression or not?
Also, exporting in your Windows share from XO as a backup could produce different result than using XS from your computer (not the same network path).
That's a lof of unknown
Normal backup, just a one-time backup. (Mainly for the purpose of importing to another XS.) I disabled compression.
OK, I will continue to test things and report back.
-
In most of cases, a continuous delta backup will be few seconds, so if you need to backup often your VM, that's a good idea.
Also, you can monitor the network speed in the stats to see what's going on. You can also show that CPU should be calm on the host why backuping without compression. The SR can be monitored because of the load average chart too.
-
Also a major difference between a backup tool on the guest: XO won't restore inside the VM. It will create a new VM (with the full file) then apply diff (if necessary).
The downside is "restore" time is higher. On the other hand, you could have lost your whole XenServer and having a brand new server, XO restore will still work.
To be able to rollback quickly, there is a solution, called snapshots.
In general, in IT operations, we use both:
- snapshots are very handy when you want to make a modification and rollback if anything goes wrong. It's also cool to schedule them every night in case
- but snapshots ARE NOT BACKUP. That's why cont. delta is perfect here: fast to create and your are protected, even in a catastrophic situation. If you lost your VDI chain/XenServer/whatever, a snapshot won't help at all.
But combining them is great. Restoration time of a catastrophic failure is not often a concern.
-
To finish, ideally we should be able to select which disk to backup with XO (e.g: you have a big data disk which is already saved with rsync or whatever). This is already possible in the XO backend, but we have to work on the UI.
-
@olivier said in BRRABill's Field Report With XenServer:
In most of cases, a continuous delta backup will be few seconds, so if you need to backup often your VM, that's a good idea.
Also, you can monitor the network speed in the stats to see what's going on. You can also show that CPU should be calm on the host why backuping without compression. The SR can be monitored because of the load average chart too.
I was doing that, and why I was confused about compression with export. Sometimes it seems like it is on, sometimes it seems like it is not. I guess it depends on what the source material on the server is.
When I get everything set back up I'll do some testing.
-
Cont. delta backup won't use compression at all.
-
@Dashrender said in BRRABill's Field Report With XenServer:
Nope.. some include some older ones.. but not all.
You can start with the newest and work backwards, refreshing the list of needed updates each time.
Just wanted to double-check to be sure this was the case, as this is what I have been doing.
I usually, actually, start with the 6.5 SP1, then move on to the latest patch, which will make some earlier ones disappear. I just keep moving backwards in time until no updates remain.
-
The reason I ask is because when I do this, XC still shows the "down yellow error" that patches are available, even though they aren't.
I saw online some ways to clear this, but wanted to make sure this was OK behavior before doing so.
-
So I figured out what the issue is.
There is a hotfix that is installed on one of my two servers and not the other.
On Server1, i went hotfix by hotfix, installing them all. On Server 2, I started with SP1, and then went to the oldest and worked backwards.
This leave htofixes installed on one that have been since depreciated.
At least that my hypothesis.
-
you think having a depreciated patch installed was causing the issue? I didn't know you were suppose to remove old patches before installing new ones?
-
You are not supposed to in any normal deployment scenario. I would not expect XS to differ in that.
-
@Dashrender said
you think having a depreciated patch installed was causing the issue? I didn't know you were suppose to remove old patches before installing new ones?
You can't. It's impossible with XS.
I do have a depreciated patch installed. So it is OK, just not needed anymore, hence why starting with SP1 on my newer installation skipped that patch.
