4

I’m after viewpoints on my situation from my fellow sysadmins. I think I have thought of the points, but want someone else’s opinion.

I have an ESXi5 Cluster consisting of 5 Physical machines: 4 in DRS Pool and 1 spare for DR. Nothing sits on this in case we need to Vmotion to spread the load during a spike we can. It's also quite handy for maintenance windows.

We have a Database server that is starting to getting more use as the business grows (A good thing!). It is gotten to the point now where I want to take the Database server out of the cluster and move it to its own separate physical hardware, because I can't give it enough virtualized resources currently. I’m of two minds whether to install ESXi onto this, as the ESXi Hypervisor has very little overhead and then I can add this machine as part of our cluster so should anything go wrong we still have another server to back this up.

On the other hand, I am thinking I could just install Windows 2008R2 and use a MSSQL Database to host this on its own as over time I think we may go down the route of database replication. There will be no local storage as this is all SAN based.

Can anyone foresee any problems with hosting this as a virtual host?

EDIT

Edit I have accepted Jeremy’s Answer as this has provoked some of my thoughts and has led me to a solution. I will be keeping this as a VM below is what I plan to do. I will be adding another ~100GB RAM into 2 of our Servers, so in case one fails there is still enough resources in the cluster. I will be setting DRS to apply affinity’s to this and allow for more vCPU’s and also network traffic and constrain this to a single host unless in failure.

5
  • So in both scenarios would the hardware specs be the same?
    – Mike B
    May 21, 2012 at 22:21
  • This might be of interest to you: serverfault.com/questions/339007/…
    – user89350
    May 21, 2012 at 22:22
  • There are organizations that have deployed almost all of their VMWare hosts with single guests simply for ease of portability. Are you planning to utilize directpath or similar to otherwise optimize I/O? These kinds of measures may limit portability and such...
    – rnxrx
    May 21, 2012 at 22:54
  • 1
    What is your current bottleneck for the database server?
    – Zoredache
    May 21, 2012 at 23:25
  • RAM ANd network IO. Already have an additional nic ind the servers for residency of storage (ISCI). Yea hardware would be the same.
    – Zapto
    May 22, 2012 at 5:33

2 Answers 2

1

After writing my mini-analysis below, I think you should keep it virtual, and add that DR host to the cluster to boot (there is no real advantage of a hot-spare in ESXi that I can really think of)

Advantages to keeping the DB server as a VM:

  • Greater availability (VM can run on any host, providing the host can provide enough resources)
  • Greater flexibility (DR is easier, backup may be easier depending on how you currently do your backups)
  • Server OS's lifetime is extended, because it is not tied to physical hardware. Of course you can P2P or P2V anything but that adds additional complexity down the road.

These reasons above are some of the standard, well-known and accepted advantages of virtualization.

Since the DB server's storage is SAN based, we can take that out of the equation for the most part.

The disadvantage to keeping it virtual is that you have to pay VMWare licensing for its resources. On a physical box you just install Windows and you are done.

If your DB server is constrained for network I/O you can upgrade the NICs in just one ESXi host and set a DRS affinity rule to run that VM on that host unless the host is unavailable.

If you are constrained for RAM, you have to add more RAM anyway.

2

This is a case of examining the database server's metrics. Do you have any performance graphs to share? Fill in some of the information about your setup.

  • What is the resource footprint of the DB server? RAM? Number of vCPUs?
  • How are you currently handling storage for the DB virtual machine? VMDK? RDM?
  • What are the specifications of your ESXi hosts? CPU type? CPU speed? RAM?
  • What else is running in the cluster?
  • What are the specifications of the storage array?

You're using VMWare Distributed Resource Scheduling (DRS), so I'm assuming the load for the vSphere cluster has been balanced effectively. Given that, what would you gain by isolating the server onto its own hardware (as a VM)? That's assuming that the resources allocated to the database VM are a small subset of what's available in the machine. The presence of DRS also means that you have the ability to give 8 vCPU and 64GB of RAM to the database VM.

Where's the bottleneck from the application side?

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .