Hi all: I am less than up-to-date on virtual machines in regards to OES servers. What are the benefits of running multiple VM servers on the same hardware rather than running one server natively with many services? On the one hand I can see efficiencies by keeping the hardware busy and that VM servers can be very specialized. On the other hand hardware failure can result in a world of hurt if all your VMs go down at once.
Thanks, Chris.
Other benefits of VMs: being able to migrate them to other systems (so
hardware maintenance does not require software downtime); I know at least
Xen and vmware can do this, and I think KVM can too.
Simple snapshots of entire systems can also be nice. Some filesystems
(notably btrfs and xfs as I recall) can also help with this, but doing an
entire system can be nice especially for…
Duplication of systems for a test environment. Properly isolate, of
course, but this is easier than cloning hardware disks unless you are
doing mirroring and then pull a disk out of the mirror, then put it into
another machine that is isolated network-wise… etc.
For the record, “containers” (LXC or the like) creates a nice middle
ground with many services and is available with SLES 11 SP2. Basically
these are insanely-lightweight virtual systems (not virtual machines as
they do not have virtualized hardware, thus the lightweight nature) and so
any services that can run this way get a smaller footprint which can be nice.
Good luck.
Chris wrote:
[color=blue]
On the other hand hardware failure can result in a world of hurt if
all your VMs go down at once.[/color]
If you mean that you /could/ have multiple corrupted servers to repair,
then I would agree with you but proper backups should mitigate that
risk. Whether all your services are on a single physical server or
spread across multiple virtual servers, the result of a hardware
failure is essentially the same.
Multiple virtual servers allow you to isolate servers/services and even
install mutually exclusive OES services on the same box.
–
Kevin Boyle - Knowledge Partner
If you find this post helpful and are logged into the web interface,
show your appreciation and click on the star below…
Chris sounds like they ‘said’:
[color=blue]
Hi all: I am less than up-to-date on virtual machines in regards to
OES servers. What are the benefits of running multiple VM servers on
the same hardware rather than running one server natively with many
services? On the one hand I can see efficiencies by keeping the
hardware busy and that VM servers can be very specialized. On the
other hand hardware failure can result in a world of hurt if all your
VMs go down at once.Thanks, Chris.[/color]
So my response to Chris’s comment is…
Many backup solutions now have the ability to back up the entire vm,
which would drastically help with an issue of hardware corruption on
the host, allowing you to restore the vms on new hardware.
–
Stevo
Thanks for the reply. Regarding hardware failure, I could use my current configuration as a good example. We have 5 OES servers as a result of a design decision made many years ago; have several servers, each doing one or two things, so that if any one server goes down we don not loss all network services. With this configuration, if one server suffers a catastrophic failure (say a backplane fails) I can maintain most other services.
Now if I were to have one VM host running five VMs so that no one VM runs more than one or two services, I have the same separation of services. However, the difference being if my host now suffers a catastrophic failure, I loss all the VMs and all services. I am not saying this is likely, but it is something I would worry about.
I do like the many other benefits of VMs pointed out in other thread responses, and they many in fact out-weigh the downside.
[color=blue][color=green][color=darkred]
KBOYLEKBOYLE@no-mx.forums.novell.com 2/19/2013 3:38 PM >>>
[/color][/color][/color]
Chris wrote:
[color=blue]
On the other hand hardware failure can result in a world of hurt if
all your VMs go down at once.
[/color]
If you mean that you /could/ have multiple corrupted servers to repair,
then I would agree with you but proper backups should mitigate that
risk. Whether all your services are on a single physical server or
spread across multiple virtual servers, the result of a hardware
failure is essentially the same.
Multiple virtual servers allow you to isolate servers/services and even
install mutually exclusive OES services on the same box.
–
Kevin Boyle - Knowledge Partner
If you find this post helpful and are logged into the web interface,
show your appreciation and click on the star below…
Yes. For this reason I believe most people host the actual VMs on a SAN
(of some sort or another) and then have a couple of VM hosts running, each
with a portion of the load. Often, like with RAID disks, losing one of
the hosts reduces capacity to a point still below the maximum capabilities
of all other hosts, so in your case if you just had two systems, each
running two or three VMs, but each able to run five if needed, you could
fire up the VMs from the crashed host on the working host until you get
the crashed host fixed. If you had eight VM hosts spreading that load is
a little easier even since now you can share more.
Anyway, that’s how I’d do it in your situation. Shared storage, and
most-importantly reliable shared storage, is necessary for this to really
work well.
I remember the first time somebody called me upset because they had lost
their entire eDirectory tree back in, probably, 2004.
Them: “We had three replicas just like Novell recommends, but we still
lost all of them.”
Me: “You had three servers all fail losing data at the same time?”
Them: “Yes, all three VMs are down.”
Me: “VMs?”
Them: “Yes, they were all on the same host which had the drive failure.”
Me: “So you distributed your redundant data across one physical drive…”
Them: “Yes.”
It’s a painful lesson to learn, but the same type of thing applies to you.
Just as you have redundant systems right now so no single machine failure
takes down your entire network, be sure you have something similar to
prevent a single failure from crippling you while virtualizing. Bosses
rarely care how great patching and cloning and similar things can be if
their bosses are telling him to fire you for letting the entire network go
down.
Good luck.
Chris wrote:
[color=blue]
Thanks for the reply. Regarding hardware failure, I could use my current
configuration as a good example. We have 5 OES servers as a result of a
design decision made many years ago; have several servers, each doing one
or two things, so that if any one server goes down we don not loss all
network services. With this configuration, if one server suffers a
catastrophic failure (say a backplane fails) I can maintain most other
services.Now if I were to have one VM host running five VMs so that no one VM runs
more than one or two services, I have the same separation of services.
However, the difference being if my host now suffers a catastrophic
failure, I loss all the VMs and all services. I am not saying this is
likely, but it is something I would worry about.I do like the many other benefits of VMs pointed out in other thread
responses, and they many in fact out-weigh the downside.
[color=green][color=darkred]KBOYLEKBOYLE@no-mx.forums.novell.com 2/19/2013 3:38 PM >>>[/color][/color]
Chris wrote:
[color=green]
On the other hand hardware failure can result in a world of hurt if
all your VMs go down at once.[/color]If you mean that you /could/ have multiple corrupted servers to repair,
then I would agree with you but proper backups should mitigate that
risk. Whether all your services are on a single physical server or
spread across multiple virtual servers, the result of a hardware
failure is essentially the same.Multiple virtual servers allow you to isolate servers/services and even
install mutually exclusive OES services on the same box.–
Kevin Boyle - Knowledge Partner
If you find this post helpful and are logged into the web interface,
show your appreciation and click on the star below…[/color]
The only way to safely run vm’s is to use High availability with a san. I
have been doing so now for quite a while and would not turn back. I prefer
to use a san for redundancy. It has been very reliable and using raw disks
off the san for vm’s I don’t see the impact that see running say off of an
ocfs disk with image files.
Just my 2 cents.
Rick
Chris wrote:
[color=blue]
We have 5 OES servers as a result of a design decision made many
years ago; have several servers, each doing one or two things, so
that if any one server goes down we don not loss all network
services. With this configuration, if one server suffers a
catastrophic failure (say a backplane fails) I can maintain most
other services.[/color]
Yes! And that’s because each server is running on its own hardware
platform. If /all/ your virtual servers run on the same box, then your
risk increases.
With five physical servers, the risk of one of them failing is five
times higher than if you only had a single server albeit only a sew
services would be affected. You would still suffer down time due to the
hardware failure.
The better solution is to run multiple virtual servers on two or more
boxes. This way, you still can isolate each virtual server and if there
is a hardware failure virtual servers can be migrated over to the other
box(s).
–
Kevin Boyle - Knowledge Partner
If you find this post helpful and are logged into the web interface,
show your appreciation and click on the star below…
Many backup solutions now have the ability to back up the entire vm,[color=blue]
which would drastically help with an issue of hardware corruption on
the host, allowing you to restore the vms on new hardware.[/color]
On the downside depending on the size of the virtual machines being backed
up, the time required and size of media for the backup are enormous.
GofBorg sounds like they ‘said’:
[color=blue]
On the downside depending on the size of the virtual machines being
backed up, the time required and size of media for the backup are
enormous.[/color]
So my response to GofBorg’s comment is…
Oh most definitely, plus the backup may be considered ‘failed’ if using
independent disks, even though the rest of the vm backs up fine.
–
Stevo
GofBorg wrote:
[color=blue][color=green]
Many backup solutions now have the ability to back up the entire vm,
which would drastically help with an issue of hardware corruption on
the host, allowing you to restore the vms on new hardware.[/color]On the downside depending on the size of the virtual machines being
backed up, the time required and size of media for the backup are
enormous.[/color]
If your backup solution and host support block tracking then it is
quick and easy.
26 guests replicated over a 10mbit link to our DR site in another city
every night is a little over 5 hours. That’s 2.2TB worth of
provisioned disk, only ~20GB gets transferred each night.
Our supporting servers tend to only transfer tens of megs each night,
it’s really only our file server and a couple of servers that do some
db backups that are large.
Our DR site has about 4 or 5TB of disk and we currently have things set
for 3 days of restore points with oodles of spare disk space left…
In addition to what others have mentioned, DR…
Virtualisation brings pretty comprehensive DR solutions within reach of
SMEs.
We purchased a server with a heap of disk and reasonable amount of RAM
for about NZD20k, dropped it in another city.
We replicate 26 guests from 3 hosts to that server every night over a
10mbit link, maintaining 3 restore points. These guests are crash
consistent - so we don’t shut down the source servers - but we can spin
them all up at our DR site within 10 - 15 minutes and have our
‘network’ back. Not ideal, but cost effective.
No need for building up the servers and restoring. Just connect to the
vSphere client, select the server and power it on.
Don’t under-estimate service isolation. Our ERP software had a number
of peripheral services that needed to be run, some from 3rd party
vendors. They tended not to play well together. Running them in their
own Windows instance has dramatically improved the stability and uptime
of those servers.
There are also HA solutions for virtual servers, software for dynamic
provisioning of servers depending on demand and the ability to move
servers without any downtime between hosts.
If your backup solution and host support block tracking then it is[color=blue]
quick and easy.[/color]
What are you using if I may inquire?
GofBorg wrote:
[color=blue]
What are you using if I may inquire?[/color]
Veeam
(http://www.veeam.com/vm-backup-recovery-replication-software.html?ad=ho
me) against vSphere 5 ESXi
Veeam will also compress the data as it replicates it across the
network, and you end up with snapshots on the remote server that you
can go back to.
There are options around backups rather than replication; backups
consuming far less space but need to be restored.
Happy to discuss further…
[color=blue]
Happy to discuss further…[/color]
Well I have <2TB of data on my network and I have a hard time fitting it
into a 12 hour window on a gigabit network. The bottleneck in my situation
are the servers being backed up some of which are quite old. I have some
servers that I can back up 100G in 20 minuts and others that take hours to
back up 4 Gig. Fortunately the slowpokes are getting replaced soon.
GofBorg wrote:
[color=blue]
Well I have <2TB of data on my network and I have a hard time fitting
it into a 12 hour window on a gigabit network. The bottleneck in my
situation are the servers being backed up some of which are quite
old. I have some servers that I can back up 100G in 20 minuts and
others that take hours to back up 4 Gig. Fortunately the slowpokes
are getting replaced soon.[/color]
We have 3 hosts and they are all under 5 years old - but my link to my
DR site is the bottleneck at 10Mbit.
The really nice thing about backup solutions like Veeam is they take
advantage of a feature of vSphere to track block changes. Veeam will
will only transfer the blocks that have changed, and can compress it
and send it to a proxy to write to a destination server. As vSphere
monitors blocks it is filesystem independent, it can even do this on
RDM luns as long as you have them set to ‘virtual’. Of-course if you
go out and run a defrag tool on your disk you’ll replicate a lot of
data.
The changed blocks end up as snapshots at the destination. So you can
revert back to previous snapshots easily. The oldest snapshot gets
merged through standard vSphere functionality.
Originally I wasn’t going to replicate my file server. It’s got a 1TB
partition with 600GB in use - we still back up regular changeable data
to tape and it goes offsite so my plan was that that server would
simply be restored from tape (most people can survive a couple of days
without access to their files, but not email, and our ERP suite).
Veeam worked so well that I replicate the server as well.
Our replicated guests also get spun up and checked once a week, so I
know that it works and works well. I’ve had one instance where a
snapshot on a replicated guest got messed up in the 12 months of
running the solution.
Chris wrote:
[color=blue]
Thanks for the reply. Regarding hardware failure, I could use my current
configuration as a good example. We have 5 OES servers as a result of a
design decision made many years ago; have several servers, each doing one
or two things, so that if any one server goes down we don not loss all
network services. With this configuration, if one server suffers a
catastrophic failure (say a backplane fails) I can maintain most other
services.Now if I were to have one VM host running five VMs so that no one VM runs
more than one or two services, I have the same separation of services.
However, the difference being if my host now suffers a catastrophic
failure, I loss all the VMs and all services. I am not saying this is
likely, but it is something I would worry about.I do like the many other benefits of VMs pointed out in other thread
responses, and they many in fact out-weigh the downside.
[color=green][color=darkred]KBOYLEKBOYLE@no-mx.forums.novell.com 2/19/2013 3:38 PM >>>[/color][/color]
Chris wrote:
[color=green]
On the other hand hardware failure can result in a world of hurt if
all your VMs go down at once.[/color]If you mean that you /could/ have multiple corrupted servers to repair,
then I would agree with you but proper backups should mitigate that
risk. Whether all your services are on a single physical server or
spread across multiple virtual servers, the result of a hardware
failure is essentially the same.Multiple virtual servers allow you to isolate servers/services and even
install mutually exclusive OES services on the same box.–
Kevin Boyle - Knowledge Partner
If you find this post helpful and are logged into the web interface,
show your appreciation and click on the star below…[/color]
The only way to safely run vm’s is to use High availability with a san. I
have been doing so now for quite a while and would not turn back. I prefer
to use a san for redundancy. It has been very reliable and using raw disks
off the san for vm’s I don’t see the impact that see running say off of an
ocfs disk with image files.
Just my 2 cents.
Rick
one thing not to forget as an advantage for VMs
Server migrations - replace your server with better/different hardware
with no driver or related issues.
We use VM even in single server instances for this reason - never ever
have another hardware dependency.
1- Install VMWare ESXi Essentials which is trivial to install
2- Migrate the VM Guest from one server to the next.
3- There is no step three. 
ashmoore’s Profile: http://forums.novell.com/member.php?userid=8630
View this thread: http://forums.novell.com/showthread.php?t=464247