|
Home > Archive > Data Storage > August 2004 > Question about Storage Virtualization ... does it make sense?
You are viewing an archived Text-only version of the thread.
To view this thread in it's original format and/or if you want to reply to
this thread please [click here]
| Author |
Question about Storage Virtualization ... does it make sense?
|
|
| Michael 2004-08-27, 5:46 pm |
| well I expect that there are a lot of very experienced and
knowledgeable people out there in this forum.
I'm wondering whether it makes sense to go for Storage Virtualization
in future.
I see a new technology that adds more complexity in my already complex
SAN. There are concepts out like inband and outband virtualization.
There is the promise of making evereything cheaper and easier to
manage.
What can I believe? Can somebiody help me with pros and cons.
Thank You
Michael
| |
|
| On 26 Aug 2004 09:51:24 -0700, mips@epost.de (Michael) wrote:
>I'm wondering whether it makes sense to go for Storage Virtualization
>in future.
>
>I see a new technology that adds more complexity in my already complex
>SAN. There are concepts out like inband and outband virtualization.
>There is the promise of making evereything cheaper and easier to
>manage.
>
>What can I believe? Can somebiody help me with pros and cons.
Virtualisation is now new and there are lots of ways to achieve it.
If you have a RAID array you are already virtualising your storage.
If you're using a volume manager you're already virtualising your
storage.
In-band and out-of-band simply refer to the topology employed by the
virtualisation device/service - it either gets in the way of your data
(in-band) or it doesn't (out-of-band).
I see the real benefits of virtualisation coming from two main areas:
- it gives you the ability to genuinely use a single management
toolset for all of your storage across all of your platforms
- you can buy any (cheap) storage from any vendor because you're no
longer looking for vendor hardware or software features. Some say
this reduces vendor lock-in.
Sure, there are other benefits, but these are the main ones.
The cons:
- You're locked in to the virtualisation vendor
- There is usually a performance penalty to pay somewhere
- It adds a layer of complexity when it comes to fixing problems
- You may lose the benefits of hardware/array features developed by
the storage vendor
Taking an extreme example...
Say you have a couple of EMC Symmetrix arrays and you're using
TimeFinder and SRDF to take volume copies/replicate data and this is
critical to your business. That's a lot of software and hardware
maintenance you're going to have to pay every year. Generally, the
older your equipment gets, the more expensive the licenses become.
If you take a virtualistion solution, you don't need the Symmetrix
anymore because you can provide a lot of the features through the
virtualisation appliance and once you do that, you can buy any storage
you like.
So... you could use say, Nexsan storage, and get the volume copy and
replication features through the virtualisation appliance. Next time
you want some storage, you can beat up Nexsan (and everybody else) on
price because you're only interested in the disk itself, not any
additional and expensive features.
Of course, I'm not suggesting that the Nexsan arrays would give you
the same performance as a pair of Symmetrix boxes (though someone here
will probably argue that the Nexsan arrays are faster!).
My main point is that the virtualisation can give you the same or
similar features at a much lower cost, freeing you from the hardware
dependence. If you have a large amount of storage you can manage it
all from one toolset, which obviously is a good thing, particularly if
you have hardware from different storage vendors.
On the other hand, maybe in the future EMC will create a
virtualisation device themselves and bring Timefinder and SRDF to the
open storage world. Now there's an idea... ;-)
I don't think the virtualisation market has hit anything like critical
mass yet - I reckon that's a couple of years away. Everybody is
fighting to play - software and hardware vendors alike.
HVB.
| |
| Faeandar 2004-08-27, 5:46 pm |
| On Thu, 26 Aug 2004 23:05:59 +0100, HVB <devnull@127.0.0.1> wrote:
>On 26 Aug 2004 09:51:24 -0700, mips@epost.de (Michael) wrote:
>
>
>Virtualisation is now new and there are lots of ways to achieve it.
>
>If you have a RAID array you are already virtualising your storage.
>If you're using a volume manager you're already virtualising your
>storage.
>
>In-band and out-of-band simply refer to the topology employed by the
>virtualisation device/service - it either gets in the way of your data
>(in-band) or it doesn't (out-of-band).
>
>I see the real benefits of virtualisation coming from two main areas:
>- it gives you the ability to genuinely use a single management
>toolset for all of your storage across all of your platforms
>- you can buy any (cheap) storage from any vendor because you're no
>longer looking for vendor hardware or software features. Some say
>this reduces vendor lock-in.
>
>Sure, there are other benefits, but these are the main ones.
>
>The cons:
>- You're locked in to the virtualisation vendor
>- There is usually a performance penalty to pay somewhere
>- It adds a layer of complexity when it comes to fixing problems
>- You may lose the benefits of hardware/array features developed by
>the storage vendor
>
>Taking an extreme example...
>
>Say you have a couple of EMC Symmetrix arrays and you're using
>TimeFinder and SRDF to take volume copies/replicate data and this is
>critical to your business. That's a lot of software and hardware
>maintenance you're going to have to pay every year. Generally, the
>older your equipment gets, the more expensive the licenses become.
>
>If you take a virtualistion solution, you don't need the Symmetrix
>anymore because you can provide a lot of the features through the
>virtualisation appliance and once you do that, you can buy any storage
>you like.
>
>So... you could use say, Nexsan storage, and get the volume copy and
>replication features through the virtualisation appliance. Next time
>you want some storage, you can beat up Nexsan (and everybody else) on
>price because you're only interested in the disk itself, not any
>additional and expensive features.
>
>Of course, I'm not suggesting that the Nexsan arrays would give you
>the same performance as a pair of Symmetrix boxes (though someone here
>will probably argue that the Nexsan arrays are faster!).
>
>My main point is that the virtualisation can give you the same or
>similar features at a much lower cost, freeing you from the hardware
>dependence. If you have a large amount of storage you can manage it
>all from one toolset, which obviously is a good thing, particularly if
>you have hardware from different storage vendors.
>
>On the other hand, maybe in the future EMC will create a
>virtualisation device themselves and bring Timefinder and SRDF to the
>open storage world. Now there's an idea... ;-)
>
>I don't think the virtualisation market has hit anything like critical
>mass yet - I reckon that's a couple of years away. Everybody is
>fighting to play - software and hardware vendors alike.
>
>HVB.
Alot of good info in this post, so the OP should take stock in it.
To continue the extreme example of inband virtualization though you've
simply moved your lock-in point from EMC to VirtVendor. In many cases
the best virtualization is done in such a way that pieces of data may
reside anywhere and you have no idea where. "Best" is always arguable
but in the spirit of virtualization that is the best. Think of it
like your car, you know there's a bunch of stuff under the hood that
makes it all work but you have no idea what/where/how (some may but
that's not the point). You are virtualizing the act of driving from
the physical layer of combustion and torque.
Same goes for data, you are writing data to something and letting it
manage where it goes. Virtualizing the act of creating data from the
physical layer of storing it.
Not that this is a bad thing necessarily but it's still new in terms
of adoption. Alot of people are skittish about this and like knowing
where their data is physically (ie. disks 5-15 hold Oracle, etc).
It's something to consider when thinking about virtualization, it
means something different to almost everyone. That makes it hard to
understand the game.
~F
| |
| Paul Repacholi 2004-08-27, 5:46 pm |
| HVB <devnull@127.0.0.1> writes:
> I see the real benefits of virtualisation coming from two main
> areas: - it gives you the ability to genuinely use a single
> management toolset for all of your storage across all of your
> platforms
Does it? As I see it it means you must now have the tools for the N
types of systems you use, plus a set for your storage thing as well,
and understand how they will interact. So you go from an order N
problem to at least an order N+1 problem.
> - you can buy any (cheap) storage from any vendor because you're no
> longer looking for vendor hardware or software features. Some say
> this reduces vendor lock-in.
Good, cheap, fast. Pick two. OR is it; Good, Cheap, Fast, Virtual,
pick two...
--
Paul Repacholi 1 Crescent Rd.,
+61 (08) 9257-1001 Kalamunda.
West Australia 6076
comp.os.vms,- The Older, Grumpier Slashdot
Raw, Cooked or Well-done, it's all half baked.
EPIC, The Architecture of the future, always has been, always will be.
| |
|
| On Fri, 27 Aug 2004 22:08:48 +0800, Paul Repacholi
<prep@prep.synonet.com> wrote:
>HVB writes:
>
>
>Does it? As I see it it means you must now have the tools for the N
>types of systems you use, plus a set for your storage thing as well,
>and understand how they will interact. So you go from an order N
>problem to at least an order N+1 problem.
Yes, I believe it does. You don't really have to manage the storage
arrays through their own tools. What you do is export ALL the LUNs to
the virtualisation device when you first set the thing up - unless you
go messing around with the array, you need never touch the vendor
tools again.
If you're real clever you'll get the supplier to configure the array
to connect to the virtualisation device for you, so you never actually
use the vendor tools.
>
>Good, cheap, fast. Pick two. OR is it; Good, Cheap, Fast, Virtual,
>pick two...
I don't think holds true anymore. There are some excellent, fast and
cheap storage products around nowadays. They aren't as fast as the
top-end EMC or HDS arrays, but they're only a tiny fraction of the
price. Lots of people get hung up on using the top of the line kit as
a benchmark when such a solution would be massive overkill for most
scenarios.
By implementing a virtualisation solution encompassing cheaper disk
arrays you can create a solution similar to the high-end kit, but
without the cost.
I'm not suggesting the two are equal. There are a lot of people who
would like high-end functionality, but can't afford the high-end
hardware to run it on.
HVB
| |
| Nik Simpson 2004-08-27, 5:46 pm |
| Paul Repacholi wrote:
> HVB <devnull@127.0.0.1> writes:
>
>
> Does it? As I see it it means you must now have the tools for the N
> types of systems you use, plus a set for your storage thing as well,
> and understand how they will interact. So you go from an order N
> problem to at least an order N+1 problem.
Most of the operations you need the storage tools to perform are one time
operations, i.e. carve out the LUNs, everyting else you can do from the
virtualization console. Several of the virtualization companies (any of them
with any sense) are looking to add the ability to do the first task through
SMI-S in the future. So yes it will largely be a single console operation
apart from initial array setup.
--
Nik Simpson
| |
|
|
|
| On 26 Aug 2004 09:51:24 -0700, mips@epost.de (Michael) wrote:
>well I expect that there are a lot of very experienced and
>knowledgeable people out there in this forum.
>
>I'm wondering whether it makes sense to go for Storage Virtualization
>in future.
>
>I see a new technology that adds more complexity in my already complex
>SAN. There are concepts out like inband and outband virtualization.
>There is the promise of making evereything cheaper and easier to
>manage.
>
>What can I believe? Can somebiody help me with pros and cons.
>
Virtualization Pros:
================
Allows transparent storage configuration across different arrays, host
doesn't care.
Can provide nearly 100% continuous access to storage when configured for
this capability (multiple virtualization servers that can fail-over)
Can standardize the management features available to all storage (e.g.,
cheaper arrays can get some of the features like snapshot & replication
that are usually only available in the more expensive arrays)
Virtualization Cons:
================
Introduces another layer of complexity in the management of storage.
Introduces another level of performance degradation between the storage and
the server
Adds another expense item that must be incurred to provide storage services
Reduces some capabilities of array-based features, though some
virtualization servers provide these capabilities as replacements
More ports used on your switches for the virtualization servers
-----------------------------------
There are other possible pros & cons. What's significant in your
individual environment may not be significant in others. A more
site-specific analysis can help to uncover the business-drivers that would
support (or eliminate) a virtualization server as a possible solution worth
pursuing in your environment. After all, if it doesn't translate into a
business value, it probably isn't worth the investment.
--- jls
The preceding message was personal opinion only.
I do not speak in any authorized capacity for anyone,
and certainly not my employer.
(get rid of the xxxz in my address to e-mail)
| |
| Nik Simpson 2004-08-30, 5:45 pm |
| jlsue wrote:
> On 26 Aug 2004 09:51:24 -0700, mips@epost.de (Michael) wrote:
>
>
> Introduces another layer of complexity in the management of storage.
>
Yes and no. As has already been discussed, the virtualization layer provieds
a common I/F for many functions like replication, volumen management etc.
The only downside is that the array tools have to be used for things like
configuring RAID stripes. So the amount of additional complexity is minimal.
> Introduces another level of performance degradation between the
> storage and the server
>
Not necessarily, if done right, the worst case (i.e. a complete cache miss
in the virtualization layer) it will add maybe 100Microseconds to the
roundtrip latency, but since a cache miss in the VM layer probably means a
cache miss in the array as well, this additional latency is negligible
compared to the overall time to retrieve data from the disks. Additonally
with caching in the VM layer, cache hits will be faster than the array in
many cases, particularly where the VM layer is being used to add additional
capabilities to low cost arrays using SATA or something like that. So the
lesson here is don't assume that adding a VM layer will have a negative
impact on performance, if implemented properly it may well have a positive
impact, and in the worst case it will not have a major negative impact.
> Adds another expense item that must be incurred to provide storage
> services
>
Not necesserily, depends what those functions would cost for the native
array. Consider as an example the requirment ot mirror a Symm to offsite
storage. Without the VM layer you need a pair of Symms + EMC replication
software for the Symms, and that ain't cheap. With the VM layer you can use
low cost storage at the remote site and yopu don't need the Symm replication
software at all. So in many cases it will work out cheaper than using the
native array functions.
> Reduces some capabilities of array-based features, though some
> virtualization servers provide these capabilities as replacements
I wouldn't consider a VM solution that didn't offer the capabilities I need,
if you don't need the advanced cpabilities like asynch replication etc, then
a VM solution that doesn't have them isn't a problem.
Just FYI for those that know me from previous posts, I don't work for
DataCore anymore, so I have no axe to grind ;-)
--
Nik Simpson
|
|
|
|
|