VMworld is just a few months away. And this year I submitted along with Eric Sloof (@esloof) a session called Snapshot Deepdive.
Snapshot deepdive
Making a snapshot of a virtual machine is one of the most common tasks an administrator can do with a virtual machine. Snapshots make our life must easier by providing a robust roll-back scenario when performing a upgrade or when we change the configuration of a virtual machine. Snapshots are quest independent and easy to use.
Snapshots can also be a real nightmare when used wrong.
Datastores can fill up resulting in a virtual machine (suspension) crash., you’re unable to delete snapshots and backup programs leaving unmanageable snapshots behind. These are all common issues that can occur when using snapshots. What about the size of a snapshot file, how large can a snapshot file grow?
This session will give you a deepdive in how snapshot actually work in the different version of vSphere and how you can solve common problems when working with snapshots.
As of today public voting is open. If you like the session proposal of Eric and myself please vote!
The public voting ends June 8.
Last weekend I implemented a new storage DRS cluster for a customer of mine. I must say, storage DRS rocks! But this is off topic for this post. If you want to know more in depth storage DRS stuff. Check Frank Denneman website.
After I implemented the storage DRS cluster I wanted to move about 60 virtual machines onto the new datastore cluster. These virtual machines where all Windows 7 virtual machines because the customer runs VMware View. So I selected the blue folder where the virtual machines resides and sorted the view on power state. First I selected the power off virtual machines and initiated a migration to the new datastore cluster. Second I selected the powered on virtual machines and also initiated a migration to the new datastore cluster. Then it was off to bed.
The next morning I check the result finding that about 20% of the virtual machines where not moved and that about 5% of the virtual machines where move according to vCenter but when I checked the datastore browser I still saw the original virtual machine home directory.
After some investigation where problem was 2 sided.
Storage vMotion doesn’t migrate all of the files that are in the virtual machine home directory.
Only files that native belong to a virtual machine are migrated to the destination. In my case the customer experimented with creating a screenshot of a virtual machine. This option makes a .png file in the virtual machine home directory. A storage vMotion doesn’t migrate the kind of file, so in the final phase of a storage vMotion the virtual machine files are deleted on the original datastore but not the ‘non-native’ file, making the deletion of the original virtual machine home directory impossible.
Because off performance reasons I can imaging that VMware only migrate ‘native’ virtual machines file. Imaging that somebody accidentally places a 8 GB DVD ISO file in the virtual machine home directory. That file is also migrated. On the other hand, maybe I wanted these kind of files in the virtual machine home directory. And know these file aren’t migrated. And how many of the VMware administrators check if the original home directory is really deleted on the original datastore? Not many I guess.
I think it would be nice it VMware changes this in the next update or release.
Power state changed while waiting for a storage migration
The virtual machines where Windows 7 virtual machines because the customer runs VMware View. The maximum total concurrent storage vMotion is 6. As I wanted to move 50 virtual machines 44 had to wait. During this wait period, the View broker powered down several virtual machines because the user logged out.
I noticed this behavior before when I use scheduled tasks in vCenter. I’m not sure why vSphere doesn’t check the power state of a virtual machine again when it started migration. For me it doesn’t matter if the machine is powered on or powered off. I just wanted it to be migrated to the destination. This makes storage vMotion a bit less usable for migrations of many virtual machines at the same time that change power state now and than.
Never the less I find storage vMotion a great tool! But in combination with VMware View……..?
Just a quick post. Last week I received a mail from VMware with information about the vExpert 2012 program. One of those items was a link to the new vExpert 2012 logo. So I thought to share this with you guy’s.
I was upgrading some VMs that were a member of a vApp.
During the process I noticed that all the VMs had really old snaphosts. As most of you probably know leaving a VM in snapshot mode for a long time isn’t wise because it can fill-up your datastore. So I removed the snaphosts from the VMs. 1 VM deleted its snaphost quickly, so I powered the VM on. During the power on stage I got the error:
The operation is not allowed in the current state
It didn’t took me long the discover the problem. The other VM was still deleting his snapshots, preventing the other VM powering up.
Though I find the behavior strange. You cannot snapshot a vApp at once and why doesn’t VMware allow me to power on a VM in a vApp when another VM is deleting his snaphosts?
While performing a performance analysis for a customer, one of the things that came up is the lack op CPU power. Meaning, He had over committed the amount of vCPU’s. This is easily resolved by adding more CPU’s to the cluster. While discussing witch hardware to buy we ended up in the following discussion.
Is it better to buy a server with a 4 core 3 Ghz CPU or with a 6 core 2.2 Ghz CPU?
In other words, what is better for performance? The total amount of Mhz in the host or the amount of CPU cores?
First let me explain my thoughts. A virtual machine with 1 vCPU can run on 1 core at a time. Although it seems that all virtual machines are running at the same time, this isn’t the fact. The CPU scheduler of the VMkernel places a virtual machine on CPU core and when this VM is done doing his thing another virtual machine can use this core. Of course when you have multiple cores in a host, more virtual machines can run at the same time.
When you have a 2 vCPU virtual machine, this virtual machine wants to ‘lock’ 2 fyscial CPU cores at the same time. When a second core ain’t available the virtual machine is co-stopped (%CSTP).
If a core has more 3 Ghz, the calculation being done on that CPU core faster than on a 2 Ghz CPU core. So a other virtual machine can be put on this CPU core for his calculation.
On the other hand, with more CPU cores in your host. More virtual machines can perform CPU calculations at the same time. Ending up in a lower %READY.
While Googling around on the internet I stumbled on this thread in the VMware Community forums. Jon Hemming explaines how Capacity planner rates a system. The first CPU core counts as 100%, the second CPU core is 90% of the first CPU core. The third CPU core is 90% of the the second CPU.
Core count |
CPU1 2 Ghz |
CPU2 3 Ghz |
1 |
2000 |
3000 |
2 |
1800 |
2700 |
3 |
1620 |
2430 |
4 |
1458 |
2187 |
5 |
1312.2 |
1968.3 |
6 |
1180.98 |
1771.47 |
7 |
1062.882 |
1594.323 |
8 |
956.5938 |
1434.8907 |
9 |
860.93442 |
|
10 |
774.840978 |
|
11 |
697.3568802 |
|
12 |
627.62119218 |
|
Total |
14351.40927038 |
17085.9837 |
As you can see, with this calculation the two 3 Ghz 4 core CPU has more power than the two 2 Ghz 6 core CPU.
Jon indicated that this calculation isn’t up-to-date for modern hardware and that the are trying to optimization this algorithm.
This post is back-upped by a Youtube movie from Intel. In this movie Intel claims that the bigger the CPU is, the more and flexibler it can run virtual machines. In this movie you will see 1 small virtual machine combined with 1 large virtual machine will better fit in a large CPU than in 2 smaller CPUs. But 2 virtual machines cannot run at the same time in a CPU? Lake of CPU cores will result in %CSTP and %READY.
So this left me a bit confused. My opinion: it’s better to have more cores than more Mhz. What’s your opinion?
I’m guessing the result will be a combination of both. But what when have the budget for more 2 Ghz CPUs or less 3 Ghz CPUs?