jump to navigation

To V or not to V? June 27, 2011

Posted by audiomatron in Virtualization.
trackback

That is the question… A question to which the answer may seem quite clear to a virtualization enthusiast. However, even for me the answer to that question can at times be uncertain, but it doesn’t have to be. Within the last month or two, I have virtualized my company’s two most critical assets – our main file server containing all of our project files, and Deltek Vision, our financial management software. I was very apprehensive about doing this, and had to very carefully consider the impact of my decision to go ahead with the virtualization of these two machines. And guess what? Since I virtualized these two boxes, my users can’t even tell the difference. So what exactly made me so hesitant about virtualizing these two servers?

The Deltek Vision Server

The vendors system requirements/recommendations on this are ridiculous. The recommendation for server hardware for this app (loosely, from memory) are three quite powerful servers – one for each tier of the app (web, SQL, and application). To me that seemed a bit steep. Granted, if the app were going to be deployed in an environment with a large user base, I could see (maybe) following their recommendations. They told me since we have a relatively small user base, I could put it all on one server.

The other thing that made me uneasy about virtualizing the app was the fact that I couldn’t get a clear answer about whether or not they supported it running on VMWare. They said something like, “We don’t support VMWare. We’ll still support the software, but not if the issue is related to VMWare.” One weekend, at a part time, one-weekend-a-month job I have, I was discussing this with a colleague of mine, Jase, who is far wiser in the ways of virtualization than I am. I told him what Deltek said, to which he replied, “I bet they don’t support Dell or HP either…” Folks, a server is a server. Your app vendor isn’t going to support your hardware (physical or virtual).

The File Server

I can already guess what you’re thinking, “dude, it’s just a file server, what are you so worried about?” This server is our single most used server. Period. We are a civil engineering company. The vast majority of our data is large AutoCAD drawings, aerial photos, maps, large PDFs, spreadsheets, Word docs, etc. This server gets hammered. Plus, the only LUN in my SAN that has enough space is on 7.2K RPM SATA drives. This server is a dual quad core machine with 10k RPM SAS drives. The last thing I want to do is to make something my users use quite often slower.

Why?

Why did I want to do this? The servers were fine the way they were, right? Well, the Vision server was of the same make and model as the hosts in my vSphere cluster, and I wanted to add another host. Plus, as I monitored the Vision server, I couldn’t help but notice that the utilization most of the time was next to nothing. It was a total waste of hardware that could be put to better use.

My reasoning behind wanting to virtualize the file server was different though. This box was running Windows Server 2003. I needed to get to a place where it would be quick and easy to upgrade the OS.

Now that we’ve looked at what might have potentially stopped me from virtualizing these servers, and also some good reasons to move ahead with virtualization, how could I be certain it would work well? How would I know that this would not cause problems for my users? The simple answer?

Just do it!

Perhaps telling you to do leap blindly into P2Ving a server is bad advice. However, simply put, that is the best way to see how well it will work.

Granted, I did my homework. I used perfmon to look at the performance metrics on these two servers. I walked around and polled my users about performance of systems that were already virtual. However, in the end, the only way to know how it is going to work is simply to go for it! Here’s how I handled virtualizing these two systems:

Over night (in the case of Vision) or a weekend (in the case of the file server), I ran VMWare vCenter converter standalone on these systems to perform a P2V (physical to virtual) conversion. I would power up the VMs that were created as a result of the process, and power down the old physical servers. I then would assign the correct IP addresses to the new virtual machines, make sure they still worked, then go to bed. I made sure to leave the old physical boxes untouched, and made sure to unplug them so that there would be no danger of them being powered back on. Most importantly, though, I made sure I had a plan in case the performance of the virtual machines was unacceptable.

On the Vision server, the main component of which is a SQL database, my plan was to leave the physical box untouched for a week so various scenarios such as payroll reports, billing invoices, and timesheets could be run on it. This would give me a good idea if it was working well. If running the app virtual turned out to have less than desirable performance, I could simply back up the database from the VM, restore it back to the physical box, and all would be well.

For the file server, I applied a similar contingency plan. If I didn’t like what I saw, I could use SyncBack Pro, a utility I use to do file backups, folder syncs, and such to copy all of the changes from the VM back to the physical box.

As it turns out, all of my planning was in vain, for you see, the two boxes perform fine, if not better as VMs. Here are some important things to consider:

User base

Remember, my focus here is small business, as I am a small business IT admin. I have 55 users. If you are a small business IT guy like me, you probably don’t have very many users either (I’d venture to say 100-200 users doesn’t even constitute “a lot” of users). Ultimately, I don’t have enough users to hammer my virtual infrastructure hard enough to make it perform badly.

Hardware

I have three hosts – two dual six core boxes and one dual quad core box each with 32GB of RAM. I have 18 VMs spanned across these three hosts. The workload isn’t heavy enough at this point to cause performance issues.

Did I mention my file server seems to perform better as a VM? At first this baffled me, but as I looked into it further, I realized some things. My original physical file server was really only making good use of one of it’s eight cores – one was constantly running high, percentage wise, and the other seven were relatively low utilization. Now, with 4 vCPUs, the load appers to be spread evenly across all cores. Plus, the cores on my ESXi hosts are nearly 800MHz faster than the ones in my old file server. Then, there’s storage to consider.

The storage that I am using for the file server on my SAN, while slower has more spindles in it’s RAID array. The physical box had four spindles, while my SAN has seven. I’ve seen it written that more spindles = better performance. According to some numbers I’ve seen, average IOPs for 10k RPM SAS are around 140, times four, that’s 560. Average IOPS for 7.2k RPM SATA are around 90, times seven, that’s 630. Realizing, of course, that these numbers aren’t absolute, it is still quite possible that my the SATA array in my SAN (half of my SAN is SAS – for databases and such, the other half is SATA) can produce more IOPs because it has more hard drives.

Conclusion

Virtualization has matured to the point where, especially in a small business setting, virtual servers can offer more flexibility and reliability than physical boxes. I couldn’t run my environment effectively or correctly without virtualization. I suppose the key things to remember when preparing to virtualize a critical system are, know your environment, know your comapny’s needs, know your users, and always have a backup plan. Keeping these things in mind will ensure a successful virtualization journey!

Marcus

P.S. – With this, or anything else I post here, if I have said something that is incorrect, or just not quite right, please feel free to correct me in the comments or via email. I’m not about leading anyone astray – I just want to share my experiences in hopes that it might help someone else. However, keep it polite please!

Advertisements

Comments»

1. Barry Le Bombe - July 1, 2011

V for Virtualization


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: