Discussion on:

17
Comments

Join the conversation!

Follow via:
RSS
Email Alert
0 Votes
+ -
Good article George, but your method depends on disk imaging technology which is unreliable and "finicky" at best. However, you are on the right track.

The key to a total no brainer success is to put the Exchange Server in a VMware GSX server virtual machine. Then you back up the VM every day, or every other day, or however often you like. The VM is a PERFECT IMAGE, not some kludge that can never be replicated in the real world. VMware creates its own "real world" by virtualizing the hardware, so you don't have to worry about the actual hardware you place the image on.

We've implemented this solution is dozens of locations and we can get a business fully recovered in less than 15 minutes after a complete meltdown. Not too bad, eh? happy

Thanks!
Tom
"Dr. ISA Server 2000"
Backing up a VMWare image is as easy as you say. You can even "Pause" the virtual server and snap away as you like. So in this respect, it is much faster to backup a VMWare image, add this to the fact that VMWare images are portable to different machines. Conceptually, it is not much different that the method I described. Most of the other recommendations on Exchange architecture and mail store design should be done in a VM environment as well.

In fairness how ever, backing up a real system image is not really that unreliable. My company has been using it for years now and it has pretty much worked flawlessly for client or server backup. The imaging technique here is important for those of us who do not virtualize our servers. Since that is the majority of Exchange servers out there, the technique described in this article caters to the masses.

I'm a big fan of VMWare for modeling and testing, but I'm not a big fan of using it in production environments for things like Exchange server as a matter of preference. The pros are obviously virtualization, but the price you pay is not just the steep licensing cost of GSX, but some performance overhead as well. I like dealing directly with the optimized drivers and hardware. Many people may prefer to use virtualization because they feel the benefits out weigh the overhead.

In my mind, $2500 for GSX server on a $4000 piece of hardware is not worth it, to others it is well justified. For some low CPU intensive things like DNS, Domain Controller, WINS, DHCP, RADIUS, Certificate Authority, and others, those make great candidates to virtualize on two GSX servers. But for other applications where the software licensing costs $10,000-$100,000, the last thing I?m going to worry about is saving a few thousand dollars on hardware.

Good to hear from you Tom
0 Votes
+ -
Good points
tshinder@... 19th Jun 2003
Hi George,

Excellent points. If you have the bucks to have the backup hardware to install the images on, and you have an established and proven routine that allows you to backup and restore the images, then I would stick with it. I'm very enamoured of how simple and quick it is to perform disaster recovery with VM's, where entire datacenters can literally be restored in less than 30 minutes.

Thanks!
Tom
Hardware is unbelievably cheap even when using SCSI based storage. Turn to IDE based storage and it's chump change.

As for an imaging server, it is no different if you use VMWare. Whether or not you store "VMDK" VMWare images or PowerQuest or Ghost disk images, you have to dump those images somewhere. Actually, disk images are compressed and compacted and are substantially smaller than VMWare images, unless you manually compact the "VMDK" file and WinZIP it. But still, VMWare images are faster to backup.

Bottom line, a server for storing disk images is dirt cheap. With 200GB IDE drives going for less than $1 per Gigabyte, hardware cost is completely negligible.
There is even newer technology out there that better complements disk imaging, and that is "Point-in-time" real time snapshots. In my experience, people don't like to down or even "Pause their VM" to take a snap shot of their system image since that takes time. More often than not, people get lazy and never bother to keep their system images up to date. This is true whether you virtualize or not.

How ever, to truly have a non-intrusive and constantly protected server, the answer is "Point-in-time" technology. In my business of selling Notebooks, "Point-in-time" technology is red hot. The ability to roll forward and back to any incremental state is an extremely effective solution. Unlike Windows XP's system restore, a utility like PowerQuest V2i Protector 2.0 Server Edition covers you even if the system does not boot. I've gone as far as deleting NTLOADER and BOOT.INI just to test if the PIT products work, and indeed they work like a charm. I already hinted at this in the article but I didn't have the space to elaborate. But the basic strategy is, take a base image and manage all deltas hands off in real time after that. No need to manually do a cold image off a hard drive or pause your VM to copy that 8 GB "vmdk" file. We all know how reliable humans are happy, because given the chance to "forget to backup that image", they will 99% of the time. Even if you don't mind coping that 8 GB file every day, you can't beat the ability to roll forward and back to any hour in time. PIT is the wave of the future for not only Data, but OS and Application as well because it make eminent sense. NetAPP uses it for data, PC makers use it to make end-user's lives easier, Windows 2003 has shadow volume, and the list goes on.

Bottom line, the key is to have a base image and maintain PIT deltas after that.
0 Votes
+ -
Point In Time
tshinder@... 19th Jun 2003
Hi George,

I've have to check out the PowerQuest V2i Protector! Sounds like a dream come true. I'll dig more into this.

Thanks again! You're one of the guys I always learn some new and cool from.

Tom
0 Votes
+ -
Thanks
georgeou 19th Jun 2003
The learning is mutual.

The basic concept of managing transactions is universal in it's application, and is not much different than a transactional database that can roll forward and back. Only in this case, read/write transactions to the hard drive are logged just as transactions to a database is logged.

You have the low end with personal computers using products like "GoBack", and on the high end EMC uses their journaling file system to replicate storage devices over great distances inaddition to providing PIT capability.

Once you start using PIT technology, you don't know how you ever lived without it. What is really impressive is that PIT technology has very little impact on performance. Never mind making a backup of a VMWare disk image in 10 minutes, making a PIT snap shot takes 2 seconds! Try beating that time . If you need to restore, hit a special key before NTLOADER begins to load and the restore application loads before the OS boots (or can?t boot). Rolling back to a certain point in time takes a less than a minute. This is because they are not really writing to the hard drive, only making a few pointer changes. Need to do a ?Bare metal restore?? No problem. It might take more than a minute, but 30 minutes is definitely doable.
0 Votes
+ -
This product is simply awesome! Very simple to use and very fast. I had been waiting for a long time for a point in time product that is a no brainier and this fits the bill. Just remember that it does not work with older OSs like NT4.

~SG
NT4 was great in it's time, but it's usefulness ran out 3 years ago. I've been on Win2k AD since 2000 and it has been great but it's time is up soon. I can't wait to start putting more things on 2003.

But in any case, you can still image an NT4server if you can't use the PIT products.
0 Votes
+ -
I think the best lesson or information that you can gain from articles like this is the clear ability of the author to think outside the box, to be a free thinker to come up with original thoughts. This article is a great example of how one admin was able to apply unique methodologies to come to up with a unique solution.

To be successful one must be able to come up with original thoughts. This is the key to success.
I really like the ideas presented here. We have not had good experience with RAID "transparency". The products mentioned do not work on Windows Server, and when running from CD or disk boot as "DOS" drivers they do NOT recognize the RAID hardware. We've tried DriveImage and TrueImage with both Promise FastTrak IDE RAID and AMI MegaRAID SCSI configurations and they consistently don't work - various problems either creating or restoring. What reasonably priced hardware configurations are people having sucess with?
I've never had any issues when I use hardware RAID. I normally use Adaptec SCSI RAID controller, but I have used the Promise Fastrak IDE RAID. Never had any problems with Ghost or DriveImage.

The point is, hardware RAID happens at the hardware level, not the driver or OS level. The OS doesn't even know it is using RAID.
0 Votes
+ -
Very timely. I just received my new copy of Ghost 7.5. We are planning a rollout of XP Pro and replacing our Netware Servers with Win2K. I was planning on doing server images of all my production systems for backup. I will use some of the tips from this article. Thanks again

Rod Gwinn
0 Votes
+ -
You're welcome
georgeou 24th Jun 2003
Although this article is titled for Exchange server, it's fundamental concepts apply to everything. The most important thing is that data should never reside on the same partition as the OS and Application because those two are the only things thatneed to be imaged. Data should be backed up with more traditional means. This not only applies to servers, but equally to client machines.

If a clients machine becomes corrupted and is performing poorly or can't boot at all, you can simply re-image the "c" partition provided all of the user's data is in a different partition.
This is the only way to fly. !!! I ghost my servers weekly and install all fixes on a test machine that is booted off a ghosted hd of actual server. Can not get any closer match to your environment than that to test updates. jtk@cablespeed.com
0 Votes
+ -
Case in point: www.techsupportforum.com/hardware-support/other-hardware-support/21229-windows-2003-server-norton-ghost-2003-problem-long.html

I tried much the same thing with DriveImage7.0 on my 2003 server w/ much the same results as in the posted link above. In the end, when i plugged in both my imaged OS & then after my original OS in the server box, i got so many NEW event errors (on both drives) that i thought that imaging wasn't possbile w/ ghost or DriveImage.

The one thing that i'm sure i did differently from most here was that i removed the physical drive from my 2003server & did the imaging on a regular workstation instead of using a bootup disk coz i didn't have the DOS version of DriveImage or a floppy on my server....but should that matter? Why would i have new event id errors when nothing on the original OS HD change. The only change really was that the server box had another os HD in it place for less than 2 hours between imaging & booting up.

I'm assuming then that Ghost12 should work w/ server2003 then? Does anyone have any experience w/ this? I just don't want my servers to be going down when i'm just doing maintainance procedures!! Could someone shed some confidence on using Ghost in the DOS mode to image 2003 servers?

thanks in advance...
I want to setup two physically identical servers exchange 2008 servers at two different sites.
I planned on using exchange replication.
So imaging the OS partition on each server will work fine. But what about the database, if I image or restore the data the second server. The other should send over the missing transaction logs right?
Keyboard Shortcuts:
Prev
Next
Toggle
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]

Join the TechRepublic Community and join the conversation! Signing-up is free and quick, Do it now, we want to hear your opinion.