VMware vSphere: Fast Track [V4] – Day 3

Another great day of knowledge transfer!
What I learned:

1. For VMware ESX, the service console always runs on the first HEC and is never migrated to another one.
2. If VMware Tools cannot be installed on a VM then the allocated CPU and memory shares should be set higher than other VMs as pagefiles or swap partitions on the VM may not always be used resulting in reduced performance. VMware Tools adds the vmmemctl also know as the balloon driver to a VM, which is used during memory overcommitment. If the vmmemctl driver is not present and part of the VMs memory needs to be taken away (e.g. if another VM has a reservation) then the memory is stolen and forced onto the vswp file for the VM. If the vmmemctl driver is present, the driver overcommits the memory on the VM allowing the VM’s OS to handle the overcommitment by moving less important information in memory to a pagefile or swap file.
3. Since by default up to 65% of memory can be paged out by the vmmemctl driver, the recommendation would be to set a 35% reservation on very important VMs to minimize, though not completely eliminate, the possibility of swapping.
4. If you see the error, “The distributed Virtual Switch corresponding to the proxy switches d5 6e 22 50 dd f2 94 7b-a6 1f b2 c2 e6 aa 0f bf on the host does not exist in vCenter or does not contain the host.” please click here for a workaround. (FYI not actually taught, but experienced during a lab.)

Clarifications I made:

1. Student Manual: As a best practice, time synchronization with the host should always be enabled.
   Comments: According to VMware KB articles this is not true: http://kb.vmware.com/kb/1318 and http://kb.vmware.com/kb/1006427.
2. Instructors: vCenter Server permissions are always most restrictive.
   Comments: While this is true if permissions are set on a group and user basis, it is not true if only set on a user basis. In the case of user basis only, permissive override restrictive and vice versa when defined deeper in the vCenter Server hierarchy.
3. Instructors: Disks on NFS datastores are always thin provisioned.
   Comments: This is because NFS is file and not block based.
4. Instructors: If a Windows VM disk is thin provisioned and a full format is performed the disk becomes thick provisioned.
   Comments: None other than to keep this in mind. To ensure this does not happen use quick format.

Questions I raised:

1. How do you check for VMs that are not registered, but still exist on datastores? (FYI I did not ask this question). They did not know, I responded by saying this can be scripted, but VMware does not provide a way.
2. If you inflate a thin provisioned disk do the fragmented files get put back together? They said inflation attempts to do this, but if space is not available then only the inflated disk space is contagious. While fragmentation cannot be determined through VMware, Storage VMotion will remove it and actually requires contagious space to be available on the destination datastore.

© 2010, Steve Flanders. All rights reserved.