Cheng's Random Thoughts on System Management : Clustering

SCVMM 2008 R2 Beta is released!

Cheng Wei — Sat, 14 Mar 2009 01:02:00 GMT

Greetings folks!

I’m pleased to announce that the public Beta of SCVMM 2008 R2 is now available for download at our connect site. Since shipping SCVMM 2008 in last October, our team has been busy working on an update that supports Windows 2008 R2, takes advantage of the new features in the platform, and addresses key customers asks. This release coming within a few months of the SCVMM 2008 RTM (including the holiday season!) delivers several exciting new features and enables a few important functionalities requested by our customers.

Here are the highlights of the new features in this Beta release:

· Live Migration support

o Windows 2008 R2 supports Live migration, and now so does SCVMM 2008 R2.

· Clustered Shared Volumes (CSV) support

o Ability to place multiple VMs on a single LUN is a key customer request.

· SAN enhancements:

o SAN migration in and out of clusters (taking advantages of your SAN infrastructure to achieve a faster migration experience)

o Multiple LUNs per single iSCSI target which provides support for iSCSI products from Network Appliance and EMC.

· Networking enhancements:

o Support for VMQ & Chimney

o Support for MAC spoofing

o Support for reusable port groups in VMWare VirtualCenter.

· Maintenance mode support

o Maintenance mode is supported across all platforms: Hyper-V, VS and VMWare ESX hosts

o Putting a host in maintenance mode, triggers evacuation of all VMs on the host via live migration on Windows 2008 R2 and VMWare.

o PRO migrations to the host are disabled.

· Disjoint domains support

o With proper permission configuration, now with VMM 2008 R2, you can easily add hosts in a disjoint domain into VMM.

For more complete description of the R2 Beta release, check out the blog post from Rakesh.

Join our connect site and download our VMM R2 Beta now!

Cheers,

Cheng

Multiple VMs per LUN

Cheng Wei — Thu, 15 Jan 2009 20:00:00 GMT

Hi there,

Rakesh just posted a new blog, through which he explains the ins and outs about the model where users create multiple VMs by using the same LUN. He also points out that, with the new Cluster Shared Volume (CSV) support in Windows Server 2008 R2, in VMM vNext (we're working on right now), users will be able to create and manage multiple VMs on a single LUN.

Enjoy your reading,

Cheng

All about refreshers ...

Cheng Wei — Thu, 02 Oct 2008 02:19:00 GMT

If you have been using VMM for a while, you’ve probably figured out that we use various “refreshers” to pull data from the hosts and VMs that we manage. When deploying VMM in various sized environments, depending on the perf/scale requirements, you might need to tweak the refresher schedule to achieve the best performance.

So, to help you understand what’s happening “under the hood”, here are the refreshers we use in SCVMM 2008 product. Please note, that not all refreshers are customizable.

Host Refresher: Runs every 30mins on every host (with time lags) and updates host properties. It also updates disk/SAN and host network (NIC / virtual switch) information. It does not check any VM related properties or perf counters on hosts. Can be manually triggered on the UI or through cmdlet.

VM Refreshers: Two types

VM Heavy Refresher: Runs every 30mins on every Host and also runs whenever a VM is clicked on (or when you call refresh-vm cmdlet). It updates all VM properties, resource pools, clustering information for this VM and snapshots. But it does not update VM performance counter info.

VM Light Refresher: Runs every 2mins on every Host that is currently in the VMM database. This refresher does the following:

- Checks the host state (to see if the agent is running or the agent is down) and virtualization software status

- Syncs the state of a VM

- Marks VMs as missing

- Imports new VMs created outside of VMM (When VMM detects a new VM created on the virtualization platform, we import it into VMM and kick off the heavy refresher for the new VMs)

Cluster Refresher: Runs every 30mins and refreshes all cluster properties (including bringing in newly added nodes or taking our removed nodes). Can be manually triggered on the UI or through cmdlet.

Library Refresher: Runs on user configurable schedule (increment by 1 hour interval and can be turned off completely) and updates the library shares info and library objects. Can be manually triggered on the UI or through cmdlet.

Perf Refresher: Runs every 9 minutes on a host or whenever there is any state changing operation on the VM (i.e. start/stop/save/etc.). It collects perf counter information of both the host and all the VM’s on a given host.

VirtualCenter Refresher: Runs every 30mins and refreshes VC properties, ESX hosts and resource pools that are managed by this VC. Can be manually triggered on the UI or through cmdlet.

User Role Refresher: Runs every 30mins and updates user role properties.

PRO Tip Refresher: Runs every 30secs. It looks for PRO specific alerts in OpsMgr and reconciles the PRO tips in our DB against the data that is brought back from OpsMgr. This refresher cannot be manually triggered.

Thanks for reading and hope this is useful,

Cheng

Clustered Hosts and HA VMs in SCVMM 2008

Cheng Wei — Tue, 10 Jun 2008 22:58:00 GMT

Why Clustering?

When IT organizations start moving their physical computing environment to a virtualized one to consolidate their servers (increase resource utilization), increase agility and adaptability of their workloads, reduce carbon footprint, or save on floor/rack space, one thing will quickly become obvious to IT admins / architects or the IT managers:

When a physical machine is hosting multiple VMs, the physical machine itself becomes the single-point-of-failure.

This also brings in the quote from Thomas Bittman, Garner Analyst, “Virtualization without good management is more dangerous than not using virtualization in the first place.”. In the all physical world, if you lose one computer, you may only lose one workload. Now, a hardware failure on a physical host can bring down multiple workloads running inside of the VMs hosted on the physical server.

Hence, how to manage highly available (HA) hosts and HA VMs naturally become one of the most important considerations in almost all customers' virtualization project plans.

What and How Can VMM Help?

Luckily, SCVMM 2008 provides a seamless experience for creating and managing HA VMs. Here are some of the highlights on what we delivered that will make your HA VM management across virtualization platforms an easy and effective one:

SCVMM 2008 is Windows 2008 Cluster "aware".

When you add a host that belongs to a windows cluster, we'll automatically bring in the rest of the nodes in the cluster.
Many SCVMM cmdlets have cluster specific parameters or switch, so that you can define how SCVMM should take actions according to the HA attribute of the target VM/host.
When the HA VM moves, we keep track of the VM and their hosting physical cluster node.

SCVMM 2008 supports both Windows cluster quick migration as well as the VMotion from VMware.
When you decide to move a HA VM, the intelligent placement takes HA hosts into consideration and will recommend clustered hosts by default.
SCVMM 2008 provides simple, actually, VERY SIMPLE, single-checkbox process to create a HA VM.

If you have tried the white paper for configuring a HA VM on Windows 2003 cluster, you know the 37 page long manual steps you have to go through to make it work.
Now in SCVMM 2008, all you need to do is to check a single checkbox, if you need a HA VM. We take care of all the configurations under the hood for you.

SCVMM 2008 offers "promotion" feature, where you can "promote" a non-HA VM to a clustered host and make it a HA VM; and it also offers "demotion" feature, where you can "demote" a HA VM to a non-clustered host and make it a non-HA VM.
SCVMM 2008 introduces the new PRO feature that allows auto-migration option when user-defined threshold is reached. This is particularly useful for HA VMs.

Below is a screenshot for where and how you create a HA VM with that single checkbox:

Here is a link where you can find definitions to some of the terms we used here. It also contains support information on the support of Virtual Server 2005 and Windows Server 2003 cluster. http://support.microsoft.com/kb/923832. Here is the step-by-step guide for creating a host clustering for Virtual Server 2005 R2: http://www.microsoft.com/downloads/details.aspx?FamilyID=09cc042b-154f-4eba-a548-89282d6eb1b3&displaylang=en

What Do I Need to Know Before Starting?

Here are the top five things you need to know before managing your Windows Failover Cluster with VMM:

SCVMM 2008 supports Windows Server 2008 Cluster. You can add hosts that are configured as Windows Server 2003 cluster into VMM. But, we're cluster-agnostic for Windows Server 2003 clusters - we'll be treating each node of the cluster as individual physical host, and treating the VMs on the host as regular VMs (instead of HA VMs). So, to obtain full benefits of HA VMs, we highly recommend you use Windows Server 2008 Cluster with VMM.
When you add a cluster to VMM, we’ll automatically add all the nodes in this cluster to VMM.

What does “add” do? We’d enable Hyper-V on the host if it’s not previously enabled, which will introduce a system reboot.
So, if you have mission critical workload running on the clusters, you may want to schedule a maintenance time before adding them into VMM.

When you migrate a HA VM, the intelligent placement will only provide ratings to clustered hosts as target host for migration. Non-clustered hosts will be listed with "zero stars".

However, it is still possible to move a HA VM to a non-clustered host. This operation is referred to as "demoting a HA VM". You will get a popup warning dialog to confirm the operation. And the VM is no longer HA when it comes to non-clustered host.

When you plan for cluster failure tolerance level, make sure to use our cluster reserve concept (or, sometimes referred to as "Over-Commit" feature) to sustain physical node failure.
In an IPSec enabled environment, you should create external virtual networks on clustered hosts before creating any HA VMs.

Failing to do so may lead to undesirable VM migration and large number of failed VM refresher jobs, due to the loss of network connectivity ==> clussvc fail on the host where the external virtual network was added ==> HA VMs on that host to failover to other host(s) of the cluster.

For more information on how to manage your HA hosts and VMs, please check out our help doc and online library.

Thanks for reading,

Cheng