In this expert videocast, backup expert W. Curtis Preston explains the major challenges to virtual server backup; the differences between vSphere's backup provisions and VMware Consolidated Backup; the various options for backing up your virtual servers; what VM-specific backup tools such as those from Vizioncore, Veeam and PHD Express bring to the backup table; how restores are different in the virtual world from in the physical world; and how dedupe and continuous data protection (CDP) impact virtual server backup.
Read the full transcript from this video below:
Virtual server backup with W. Curtis Preston
Dave Raffo: Hello, and welcome to a SearchStorage.com
expert videocast on backup for virtual servers. My
name is Dave Raffo, and I am the senior news director
for the TechTarget Storage Media Group. Joining me
to discuss backup for virtual servers is Curtis Preston.
Curtis is an executive editor at TechTarget and an
independent backup expert. Curtis is also a frequent
Storage Decisions speaker and is widely regarded
as one of the foremost experts on data backup.
Welcome, Curtis. What are the main challenges
of backing up virtual servers?
W. Curtis Preston: There is really only one challenge,
and that is the laws of physics. When you put 20
physical servers into one physical server, and you
pretend like there are actually 20 physical servers,
you very quickly run into e=mc2 and a whole bunch
of other laws. You find out that you are not going
to get your backups done in a reasonable amount
Raffo: For VMware users, is vSphere an improvement
over [VMware] Consolidated Backup, and how are they
Preston: That is like asking, Is Mac OS an
improvement over DOS? I would say yes. vSphere
solves two of the main problems that VCB has,
the first being that you have to do an image-level copy.
You have to copy over the entire VMDK file every time.
Then you copy it again, so you do this two-step backup
and you do a two-step restore. When you do a restore,
you have to restore it back to the proxy server, then it
has to import that file back to the ESX host. That's the
first problem that it solves.
The second problem is that in order to do an incremental
backup … most of the products have to … copy that full
VMDK file again, then look at that file to figure out what's
changed. What that means is, an incremental -- from an impact
on ESX [standpoint] -- has the I/O equivalent of a full,
so you have those two problems. vSphere gets rid of
both of them, so you can have a single-pass backup
directly from the VMDK file, directly out to whatever
you are using for storage for your backup system,
and it does Changed Block Tracking, so when it's
time to do an incremental backup, you can simply
ask VMware, ‘Which blocks have changed?’ and it
will say, ‘These blocks have changed,’ and the magic
Raffo: What are the options for backing up your virtual
Preston: There are several. The first, and most
common, is to just stick your head in the sand and pretend
that [your servers] are not virtual servers, and I don't
mean that in a negative way.
[The option is to] pretend there are physical servers
and put in clients, put in agents, all the stuff that you
would typically do with physical servers. The problem
there, as we already talked about, is physics; you do
run into some walls that you have to figure out. The other
[option] is VCB and all of the products that work with VCB.
There are point solutions like vRanger Pro, Veeam, esXpress;
these are products that do nothing but back up VMware.
I would also suggest that … products like source dedupe
software and CDP -- since those [take] an incremental
forever approach -- can help with the laws of physics.
It is still the original idea, where it's backup software
that is pretending it is a physical box. Since it's
incremental forever, it has a much lower I/O impact
on the server.
[The final option] is to put the VMware
data, the VMDK files themselves, on a piece of storage
that understands VMware and can do backup-related
things. The products that do that today are [from]
NetApp, EqualLogic and FalconStor. They all have
a way of taking a snapshot, but taking it in a way that
makes VMware happy, so doing whatever magic they
need to do with VMware. Each of those approaches it
completely differently, but they do some magic in
VMware, then they take their snapshot, then you can
replicate that snapshot. Now you have an actual backup,
and it has very little I/O impact on VMware, and it's
totally supported. The other storage vendors are
working on it, but those are the only three that I know
that have that today.
Raffo: What do applications like Vizioncore's vRanger
Pro and PHD's esXpress do to help backups with
Preston: This is sort of the argument of the purpose-built
box or the purpose-built application. … The only thing
[these applications] care about is VMware, and other
virtual products, like Hyper-V, but this is all they do,
and they aren't tied to VCB, for example, so they can
either do their backups with or without VCB. Since
a lot of people have found as many problems with VCB
as they did help, that is a good thing. They are going
to have much more granular restores and interesting
features that are going to apply only to VMware.
There are so many features that a large company
like Symantec or IBM or EMC are going to add
to their products just for VMware, given that it is
only part of their problem. They all do the same thing. …
It is a VMware-centric approach that people, if they
have a VMware centric-view of the world … end up liking …
quite a bit.
Raffo: Are restores different when backing up with
Preston: That, again, depends entirely on the approach
that you take, but if what we're talking
about is VCB … the restore was different, in that you
had to move the data twice. [With] all the other options,
the main thing that you need to keep in mind when you
are doing a restore is that, again, you are restoring to a
virtual server and you could have an impact on the
performance of other production VMs on a machine
when you are restoring a completely unrelated VM.
That is about the only thing I can think of that you
really need to think of there.
Raffo: How does data deduplication fit in with
Preston: Dedupe fits in a lot of ways with virtual
servers. The first that comes to mind is that virtual
servers have a higher application-to-data ratio.
Let me explain what I mean by that.
A large database server might have 5 or 6 GB of OS
and application data. Then it has, maybe, 10 TB,
maybe even more, of the data that goes with the database.
It's a very small ratio. VMs tend to be much closer;
for example, you might have 5 or 6 GB of OS and
application files, and maybe 10 GB of data.
Why do I say that? What's important there?
What it means is that a significant portion of VMs,
20 different VMs sitting on one system, are exactly
the same files. If you have got a system that can
dedupe that data on the primary side without a
performance hit, then you have got a real use case there.
That is the first thing that I think of.
The second thing has to do with [the fact that] a
lot of the different ways that we back up VMware
create more duplicate data than a traditional method.
If you are using VCB, if you are using vRanger Pro, if
you are using any of these products that want to make a
disk copy, then you end up doing a lot of full copies, more
so than a traditional backup paradigm. You end up creating
even more duplicate data than a traditional backup system would.
Therefore, that's a perfect match for deduplication.
Third, of course, is that source dedupe is one of the great ways to back
up VMware, because it is an incremental forever approach.
Raffo: How does CDP work with virtual servers?
Preston: CDP is another one of the incremental forever
approaches. It is a block-level incremental approach where
… every time a byte changes, that byte is then immediately
copied out to the backup system, and it's just happening
throughout the day. The story that I am told -- I have never
actually verified this statement -- [by] the CDP vendors is
that the I/O impact of CDP on a server is roughly that of a
virus protection software, that it just has to take the file and
then take the block as it changes, and send it out to another
location. Of course, CDP has some really cool things about it,
about being able to recover right up to the point of failure and
all those other interesting things that make CDP what it is,
but it also is a really interesting way to back up VMware.
Raffo: Thank you, Curtis. That wraps up today's expert
videocast on backing up virtual servers, with Curtis Preston.
hanks for joining us.