Virtually Speaking

Dan Kusnetzky, Paula Rooney and Ken Hess

Don't Throw Away Your Physical Servers Just Yet

By | June 8, 2011, 3:00am PDT

Summary: You’ll need to keep some physical systems around for those workloads that can’t go virtual. And, be sure to keep a horse and buggy around when that whole automobile thing doesn’t work out too.

Oops! Did you get rid of ALL your physical systems so that you could convert every workload in your data center to a virtual one? Yeah, you might want to rethink that one and go grab a few physicals just to tide you over until virtualization takes a bigger technological leap. Yes, I am a virtualization advocate. Yes, I am a cloud advocate. Yes, I am a virtualization writer. Yes, I work with virtual machines and hosts on a daily basis. And, yes, I’m a realist. Some would go so far to say that I’m a pessimist.

I’m not pessimistic about virtualization, the cloud or server consolidation. But, I am pessimistic about your company’s ability to wean itself off of physical systems.

I hear and read too many complaints about virtualization performance, legacy systems and blah, blah, blah–so much so that I’m saying, OK, maybe you should just keep some of your damn physical systems on hand to make yourself feel better. Because as everyone knows, or should know, there are just some workloads that can’t be virtualized. I know you have one that absolutely can’t be virtualized because it’s different or special in some unique way that just wouldn’t lend itself to virtualization.

Uh, OK. Whatever.

VMware begs to disagree on that point and they would be the one to know. But, hey, your non-technical executives and project managers know more than the people who brought server virtualization to the Data Center.

So, please–pretty please, don’t throw away your physical servers just yet, just in case you have some workload that can’t be virtualized.

That legacy application that should have been updated eight years ago or that Windows NT 4.0 system that you just can’t take away must remain physical. It’s OK, I’ve heard it all before. You’re absolved of any wrongdoing or mismanagement. After all, I don’t have to pay your bills. Or listen to the moans and groans of keeping that old garbage alive. And, it’s a critical piece of garbage that your business depends on. Good job on that. You deserve a promotion for your proactive work, your acute business insight and your ability to keep that mission-critical up and running no matter what the cost.

And, you had better be glad you don’t work for me.

My best advice, other than keeping those physical systems around, is to engage VMware, Citrix, the Open Virtualization Alliance or Fred the Virtualizer down the street to give you some direction in bringing your application up to date and migrating it to virtual architecture before it flops so hard that even Dr. Oz can’t revive it.

So, talk back and tell me, are you still keeping physical systems around for those workloads that you just can’t virtualize? I’d also love to know if you’ve engaged one of the companies that can help you move that workload to a virtual environment and the outcome.

Kick off your day with ZDNet's daily e-mail newsletter. It's the freshest tech news and opinion, served hot. Get it.

More from “Virtually Speaking”

Topics

Kenneth 'Ken' Hess is a full-time Windows and Linux system administrator with over 15 years of experience with Mac, Linux, UNIX, and Windows systems in large multi-data center environments.

Disclosure

Ken Hess

My full-time employer is EDS (HP). I write as a freelancer for ZDNet. The postings and opinions on this blog are my own and don't necessarily represent EDS's, HP's, their subsidiaries or affiliates positions, strategies or opinions. I own no investments or direct financial instruments in the companies I write about.

Biography

Ken Hess

Kenneth 'Ken' Hess is a full-time Windows and Linux system administrator with over 15 years of experience with Mac, Linux, UNIX, and Windows systems in large multi-data center environments.

Ken writes on a variety of topics including interoperability, virtualization, data center operations, databases, and open source software. He has written and co-written books on Linux, databases, and virtualization. He currently writes a System Administration column for Linux Magazine and is a regular contributor to Linux User & Developer magazine, ServerWatch.com's Trends and InfoStor. He often contributes to other online and print publications as well.

His first computer was a Commodore VIC-20, which he purchased because William Shatner was in the commercials.

In his limited spare time, Ken enjoys painting, drawing, and flinging angry birds at fortified pigs.

Related Discussions on TechRepublic

Did you know you can take part in these discussions with your ZDNet membership?
57
Comments

Join the conversation!

Just In

RE: Don't Throw Away Your Physical Servers Just Yet
filhomarques 16th Jul
@fr_gough

Where does Idaho rank? We have been living in Montana for the past 5 years and I am not supri sexy shop to find it #3 on the "worst" list. Considering a sexshopmove to Idaho to escapthe high cost of living a low income in MT. There may not be a sales tax here but they get you if you own property!
0 Votes
+ -
The fact that you get all in a lather because some company doesn't want to virtualize says more about you than it does them.
0 Votes
+ -
Contributr
@fr_gough

It's more of a tizzie or hissy fit than a full-blown lather.
0 Votes
+ -
@khess

More puerile than that, I think it was just a bitching session because he missed out on a contract.
0 Votes
+ -
@fr_gough I was thinking the same thing. I felt like I was talking to my 17 year old daughter having a hissy fit as I read this. But you know I'm sure he's an expert on ALL systems in ALL enterprises and knows exactly that everything can be vitalized.

- Tell you what we are using (beta testing) Citrix Receiver to bring in our legacy apps that cannot be moved the cloud and to use office, becasue Google docs does not quite cut it yet. That slow response (inside the company) is a fignament of our imagination and we can ignore it...
@fr_gough

Where does Idaho rank? We have been living in Montana for the past 5 years and I am not supri sexy shop to find it #3 on the "worst" list. Considering a sexshopmove to Idaho to escapthe high cost of living a low income in MT. There may not be a sales tax here but they get you if you own property!
@fr_gough

Where does Idaho rank? We have been living in Montana for the past 5 years and I am not supri sexy shop to find it #3 on the "worst" list. Considering a sexshopmove to Idaho to escapthe high cost of living a low income in MT. There may not be a sales tax here but they get you if you own property!
I keep one physical server and that is my router with my phone system and kvm-based (libvirt) virtual machines. As for performance? No problem with hardware virtualization. But since I have an AMD 880G-based motherboard, if only my motherboard could support IOMMU, so that I can use my physical TV tuner in my MythTV-based VM as a backend server. My MythTV backend is not in use since I have trouble with my antenna getting signals from TV transmitters about 30 miles away, even with CM-4221HD, a replacement for CM-4221 that is overly bent (32304, across from Tallahassee Community College which I'm about to move away from).
0 Votes
+ -
I've virtualized very little of my servers. Development and QA are done easily enough, but, in testing, VM latency kills our public web-hosting applications under peak load. I think it will be a couple of more years, before it will possible to fully virtualize our production environment. Even then, databases and other large memory applications should probably stay on physical hardware. Honestly, I don't see the point of virtualizing fully busy systems. It doesn't save anything because you have to have the redundancy to ensure your applications stay up in the event of a hardware failures. I expect that a CPU core count continues to grow it will make more sense, but in the mean time, I plan to deliver top performance through the use of physical servers.
0 Votes
+ -
"VM latency?"
Bit-Smacker 9th Jun
@ncted
I don't understand how you could have VM latency. In EVERY case, at my place of work, a virtual server is faster and more responsive, even when virtualized on top of its own original hardware.

Unless... You may be trying to use that garbage from MS. Try VMware, even the free version, and you can't help but be impressed.

For disaster recovery, we individually image each VM with Acronis each night. If something bad happens, we can use VMware Converter to directly convert the Acronis backup image into a running and fully-functional VMware virtual machine. I can't imagine ever wanting to run a dedicated, OS-on-metal, server ever again. I even use the same combination on my home (hobby) server (PC), which is a cheap 3-core AMD, and it's running 5 game-server VMs a full capacity most nights.

Virtualization is the ONLY way, in my opinion. The only ones who would probably disagree are the ones trying to sell you more low-end hardware.
@BitSmacker
"a virtual server is faster and more responsive"
Pull the other one...
@BitSmacker

I'm guessing that you work for VMWare as your post sounds like a copy and pasted VMWare website testimonial. Seriously.

I've tried a few virtual machine packages and always have seen some small amount of latency. And I've tried both Windows and Linux as hosts for those packages.
0 Votes
+ -
Contributr
The use of virtual machine software, that you've described as "virtualization," is not the correct approach to solve many IT problems. That is a major part of the reason that companies have not moved everything into a virtual machine.

I have trouble with the premise of this whole post. happy

Dan K
0 Votes
+ -
Exececutives vs VMware...
wzrobin Updated - 8th Jun
"VMware begs to disagree on that point and they would be the one to know. But, hey, your non-technical executives and project managers know more than the people who brought server virtualization to the Data Center."

While obviously there is no argument on the technical side with your statement, it is a little one sided...

VMware's main motivation is to sell virtualization solutions... so they have a strong financial motivation to take an optimistic view...

A potential client company's execs (and really all of it's employees) job is to protect the company, so it's their duty to look at risks and worst case scenarios...

The Execs would not be doing their jobs if they took salesmen (VMware) at their word on the product they are trying to sell, when it comes to mission critical functionalities for the organization.
0 Votes
+ -
Sure, there are applications that require sub millisecond transaction speeds. Our shop floor systems run Real Time Operating Systems and their supporting applications that generate literally millions of transactions in a short amount of time.

Not too fond of virtualizing Active Directory Domain Controllers either, but that is a risk perception issue than a hard limit. I like my domain controllers to always be the first applications up and running after a system-wide shutdown.

Many SQL loads for decision support systems virtualize just fine. Low volume OLTP as well.

As long as you invest the money on good VM hosts (i.e. triple digit RAM and multi-proc, multi-core) you can virtualize quite a bit. There are workloads that can't, sure. But without a good investment in VM hosts, and VM software everything comes crashing down.
0 Votes
+ -
Virtualized DCs
crazydanr@... 8th Jun
@facebook@...
I was hesistant as well, but we did both of ours a couple of years ago (I used one physical and one virtual for a while to test) and it's been running great. We keep a decent extra capacity of resources, and every once in awhile I manually move the DC to a server that has a few less VMs than the others.

It's scary at first, but you'll find it works pretty flawlessly on a properly set up VM environment, which it sounds like you have a good understanding of.
It should be the Pareto Principle with 80% virtual and 20% physical. The physical machines should be running those applications that are fully utilizing the hardware, that have some proprietary hardware dependence, etc. Definitely have DR done via virtualization. In any case, the application should be analyzed for candidacy for virtualization no matter what.
0 Votes
+ -
Contributr
@justthinking

That's a pretty good plan, actually.
0 Votes
+ -
Contributr
@OffsideInVancouver

I didn't miss out on anything.
Asking VMware if everything should be virtualized is like asking the fox if the door to the henhouse should be left open...
0 Votes
+ -
In most cases that's not an option that's in my hands. I have customers that I suggest should update their critical apps, and who agree with me, but the vendor selling the app won't get off the dime.

Today I'm dealing with an app that "requires" XP. I have begged the vendor to please make this native to win 7 AND port a client to Linux.

All we get are the occasional "updates" that are little more than shoving buttons around on the screen, probably to make the appearance of change. In reality it just angers the poor end users. (who btw tell me some functionality has actually been lost)

I have to deal with more than one application that from all observation must have spontaneously assembled itself and emerged from a stone wall.

I deal with one vendor that shunts you to an answering machine every call. The ONLY calls they return are when you tell them you need to buy another license. And the call back is almost immediate, so you know someone is sitting there screening it.

btw the app here will not run properly on vista in compatibility mode, but does run satisfactorily on win 7 in XP mode. Thank God.
0 Votes
+ -
@pgit Good thing you don't work for him...
@pgit If your and your clients are depending on that sort of crap you had better be finding some options while you can.
While Virtualization seems to be a nice tidy package to keep all of our servers up and running, I find the costs overwhelming. I have to keep buying VM Host solutions in the range of 30K+ when the previous server's value is no more than 3K on Current prices. I can get more than 1 physical server virtualized on a host, but I definitely can't get 10 without having to go to a bigger faster more expensive VM Hosts. The more I virtualize, the more I spend in buying virtualized equipment, not to mention the more difficult it is to get the Powers that be to sign off on another 30K server. The cheaper servers get an automatic signature and dont have the VM licensing costs either. Besides the power savings what am I missing?
0 Votes
+ -
@MIS@...
What you have said is what I was thinking. The price of good servers, RAM, and storage are at an all time low. Since you need to pay for the virtualization software and the software installed inside each virtual machine, and you need an expensive high end server to run it all on, how are you saving anything?

And if you do have everything running on one giant server, and it goes down, does everyone get the day off with pay, except you? If you have 4 critical packages running on 4 physical servers, and one goes down, you are still 75% operational. If your one big server goes down, you may as well let the customer's calls go to voice mail.
@michaellashinsky@...
You will most likely have multiple VM hosts so you spread the load around to all of the VM hosts for load balancing and redundency. If one host goes down you move any guest VMs to another host and they are up and running with only a short downtime (time it takes to boot). So in your senario your uptime would be better then 4 physicals, because a single failure would not mean a 25% reduction in operations for a day it would mean a 25% reduction for only a few minutes (However long it takes for the VM to boot on another host). As for cost the ROI is very quick. If you are a windows shop you will save tons on windows licensing as you can purchase a single license to cover your physical processors yet run many copies of windows. If you had each of the servers physical you would need to purchase a license for each server. Also I am not sure why people would spend 30k on a single server, we spend around 15k and that gets us enough processor and ram to run 10 to 15 virtual servers.
Not saying these are good reasons, but here is some of the things that kept upper mgmt from approving virtualization products:
Startup cost too high.
Application vendor says virtualization not supported.
Keep it on physical "until we go to the next version"
"Server" Virtualization offers and delivers far more effecient computing than physical servers but thats not to say that its right for every workload. I have personally virtualized scores of production systems including migrating physical to virtual and building net new systems in Xen, VMware and HyperV. Of these I have Virtualized AD Domain controllers, SQL Servers, Exchange servers, Citrix and MS TS Servers, Web Servers, Windows and Linux Servers and Desktops with amazing results. My feeling is that 80 to 90% of the existing physical servers in small to medium businesses can and should be virtualized. Some heavy workloads SHOULD NOT be virtualized and in almost ALL of my virtualization projects there is double or triple redundancy to insure systems can suffer some faults and continue to run. There are very good options for low cost server virtualization. My company runs 17 windows and linux servers on two XenServers with ease, they have been up and running for four years with maybe one or two minor limited outages.
0 Votes
+ -
@moosesg

Forgive my ignorance, but how can a virtual machine be faster than the physical machine it resides on? It would seem to me that if the advantage is running in a virtual RAM environment, couldn't you make a virtual RAM drive on a server and get similar results?
Anyone who's keeping NT boxes around should be shot for the following reasons:

1. It's no longer supported (sort of) - so BSOD really means dead.
2. It's running on a system w/o a warranty.
3. The person who wrote the legacy app is probably gonzo.
4. You are seriously behind the times.
5. There's an app for that (apologies to Apple reference).

Any one of those reasons is enough to garner support for retiring any application that's long in the tooth. Chances are most if not all the reasons make it an imperative to mothball the dinosaur. Support on the OS exacts a heavy toll and the drives are probably WAYYYYY past their drop-dead date.

If you can't resurrect the application in the event of a flameout, it's my strong suggestion you find another application to replace it. Or become very good at writing CYA e-mails and keep them handy when the clunker dies for good and you're called onto the carpet. At least then you can break out the "I told you so" and save your job happy
0 Votes
+ -
Contributr
@mohara000039

Exactly. It seems that people think that virtualization is the problem. Really? You have an app that runs on an OS that's more than 10 years old and you think that virtualization is the problem. I have some beach front property in Arizona....
I guess I should virtualize Premier and do HD video editing in the cloud.... I'm sure it'll not take too long to transfer 10GB of file up to the cloud for editing..... I can see it all now, should be super efficient.
0 Votes
+ -
Contributr
@bwalker

You might be surprised.
"Did you get rid of ALL your physical systems so that you could convert every workload in your data center to a virtual one"

If you did, your "virtual" center would cease to exist. Virtualization isn't magic; it runs on top of physical machines.
0 Votes
+ -
Contributr
@CobraA1

Yes, but I'm referring to physical systems that run single OSs. But, you knew that.
@khess Then make it clear. Do not lie.

I absolutely abhor this practice of pretending there is no physical infrastructure when people are discussing virtualization or cloud computing.
Somebody needs to get a life, or at best, a new job, because, the above article/blog, is a complete waste of people's time.

I expected to read something insightful about the alternatives to servers or desktops, but, all I found in the blog was fluff.

I hate that I wasted time even writing a response to the above, but, stuff like that should never be allowed to make it into the internet.
0 Votes
+ -
Contributr
@adornoe@...

I'll have some technical stuff up for the next few days. Maybe that will tickle your fancy.
BTW, it takes a lot to tickle my fancy. You don't want to get on my wrong side, so you better be good. wink
This is one of the dumbest articles on ZDnet yet. I could rip it apart in dozens of ways, but why bother?
0 Votes
+ -
Contributr
@thombone

You have a piece of virtual chalk, please, entertain me.
0 Votes
+ -
A while ago...
UrNotPayingAttention Updated - 8th Jun
I read where the recommendation was to virtualize everything, except leave Physical boxes for the Vcenter server, and if your virtualizing Active Directory, leave a physical box for a domain controller that has the PDC role.

Any thoughts from qualified people whether this is still a best practice?

There is a bit of a conflict of interest when VMware is telling we should virtualize everything (jus' sayin)

edit: here is the link I referred to
http://www.windowsitpro.com/article/active-directory/virtualizing-active-directory
0 Votes
+ -
Contributr
@chmod 777

Yes, there are some people who say that you should leave one domain controller as physical and you can virtualize the rest. I am not sure why this would be the case. There's nothing particularly special about a DC so not sure why it couldn't be virtual. There's a lot of paranoia around virtualization. People kept riding horses for a while after the horseless carriage came out too. It takes time.
0 Votes
+ -
Well, not just a domain controller...
UrNotPayingAttention Updated - 8th Jun
@khess

The DC that holds the PDC role is arguably the most important... it synchs time across the domain (forest), it is the master that sends out updates of Password changes, authentication failures, accounts lockouts; and when a group policy object is changed, the policy is changed on the PDC role holder, then updated to all other DC's.

I can see the positives to both sides of that argument... I can understand the reasoning behind virtualizing that one DC, but... if you're SAN goes bye-bye, I can understand at least having a (somewhat) functional domain to log in to.

I'm sorry, but to say "there's nothing particularly special about a DC" is incorrect... the DC's that hold FSMO roles are of some significance. Yes, those FSMO roles can be forcefully acquired, but that does carry risk to the A/D environment.

edit: here is the link I referred to
http://www.windowsitpro.com/article/active-directory/virtualizing-active-directory
0 Votes
+ -
Contributr
@chmod777

SAN goes bye-bye? That would take some doing. If you have a development or test environment, try it there and see how it works.
0 Votes
+ -
"That would take some doing"...
UrNotPayingAttention Updated - 8th Jun
@khess

Not as much as you would think...

I personally have experienced ( painfully ) that a x.x.1 difference in firmware between 2 SAN units (identical models, same manufacturer, same vendor, puchased at the same time) will cause both SAN units to freeze/hang upon trying to perform a rudimentary scheduled snapshot synch of 1 virtual server between the SANs. And that happened on any scheduled snapshot synchs of all the virtual servers that we migrated in our environment.

Mind you, that x.x.1 difference in firmware between the 2 SAN units was shipped to us that way from the manufacturer.

All data, all VM hosts, everything... inaccessible. Until you can hard boot both SAN devices.

Had we migrated all of our domain controllers, we would have lost our functional domain as well as our app servers/ data. (until the issue was resolved)

But... I agree with you about a test environment. As the saying goes: "First we try, then we trust"
0 Votes
+ -
Everything said
jeffh0821@... 8th Jun
Ken, I think you just had the proverbial smackdown, but in case you didn't actually learn from something from this. Let's take one of your throw off sentences.

"And, you had better be glad you don?t work for me."

If you worked for ME, you would have been fired already. You beleive one size fits all. You drink more Kool-Aid than an ADD pre-teen. You thought client/server was the end of computing. You had 20 websites that did nothing during the .com boom. You outsourced, then insourced, then co-sourced, and are now cloud-sourcing.

You are a consultants dream.

You have never generated a dollar of revenue and assume every company you have ever worked for is an IT company.

Problem is - your days are numbered. Good luck when the 20-year old Chinese graduate student, who is smarter and cheaper than you, takes your job.

Never, NEVER, lead with your chin.
0 Votes
+ -
Contributr
@jeffh0821@...

Damn, I suck.

I've actually never done any of those things but you do have a knack for interesting, albeit disjointed, dialog.

You need to leave the basement you live in once in a while and get some fresh air. Put the cases back on your computers, log out of WoW, put down the joystick, take a shower and go enjoy some quality time with a dog or a live person. You're taking this way too seriously.
Your VMs have to run on a physical server somewhere. There will be fewer of them, but of course you will have to retain some physical servers, or buy new ones that can run more VMs. Are you talking about the alternative of using a cloud service rather than running your VMs on your own physical servers?

This may have been addressed in the comments, but I don't get where you are coming from based on the article.
0 Votes
+ -
Contributr
@sam_sawyer@...

I mean physical servers in addition to the hosts.
0 Votes
+ -
Why i virtualize...
stonebit 9th Jun
Because snapshots help with those pesky unrepairable Windows crashes. Seriously. I've never met a Windows system that didn't crash within a 90 day window. I've moved a few SMBs to SAMBA4 DCs and put these on VMs. It's super. I already had to upgrade hardware, so moving the VM was a snap. The DC comes up, then the rest are brought up 1-5 minutes later (thanks ESX). I love that after an outage all the servers come up in the right order. We've even converted physicals to VMs with some of those fancy tools. They take hours, but that crappy system stays up in a virtual environment longer than it would on old hardware. And web servers? We've got about a dozen running nginx. It's very easy on the resources and has been up for 5 months (since setup). All of this runs on a big SAN and a stack of blades. And the blades power on and off depending on time of day and percent load of the lot all by themselves. Gravy train.

Join the conversation!

Formatting +
BB Codes - Note: HTML is not supported in forums
  • [b] Bold [/b]
  • [i] Italic [/i]
  • [u] Underline [/u]
  • [s] Strikethrough [/s]
  • [q] "Quote" [/q]
  • [ol][*] 1. Ordered List [/ol]
  • [ul][*] · Unordered List [/ul]
  • [pre] Preformat [/pre]
  • [quote] "Blockquote" [/quote]
ie8 fix

The best of ZDNet, delivered

ZDNet Newsletters

Get the best of ZDNet delivered straight to your inbox

Facebook Activity

White Papers, Webcasts, & Resources
ie8 fix