VMware is set to turn storage administrators into the energy-profligate and hated SUV drivers of the corporate world, a top analyst has claimed. That's because consolidation will cut servers' share of data centre power consumption and propel storage to pole position in IT's energy greed league, claimed Enterprise Strategy Group …
SAN is better?
Really, SAN is better is it? Add another 40 Watts per fibre channel port, another few kilowatts per cabinet of jumped up Linux boxes that you spent £10k each for because it was called a 'data mover'.
And as for MAID, most of the power cost in a current corporate data centre is the fixed overheads of the building, controlled by the power provisioned to the kit, so even ignoring the issues of heavily virtualised storage keeping those 'idle' disks spun up, the massive power supply ratings of MAID systems mean they will chew through the power, spun up or not. If you want some greener storage consider internal SAS disks that actually shut down when a server is idle or maybe a nice big optical or tape storage array at a fraction of the power requirement.
Next time you are looking for storage and the vendor's green paint is still not dry, before buying the claims about reduced numbers of disks in their expensive box remember the cost / performance / energy argument is two sided for SANs and count up the energy and financial costs of all the 'junk that isn't disk' as well.
VMware creates massive power savings, this means that storage gets worse? Eh? That's like saying that someone stopping using a fleet of SUVs and starting to use a bus makes someone using a truck use more fuel. Or something...
Still it's good news about not re-using all of the existing DAS disk. I could do with some nice fast disks at home...
Seriously, I don't think it'll be long before storage power usage drops significantly. The first to go will probably be backup to (permanently powered) disk, with a return to nice low power, low heat tape systems, or robot loading disk carts which are beginning to emerge.
Office environments often have a zillion PCs networked to one poor overworked server with a piddling sub-TB. Everyone complains that there's not enough storage, while those zillion PCs' often-250GB C: drives spin their lives away, 90% empty.
In this age of gigabit-Ethernet, someone should create a distributed application that will take that office full of near-empty 250GB hard drives and make a new type of multi-TB Torrent-like RAID.
One could arrange so much redundancy (something like 10:1) that failures wouldn't even be a significant concern. One could probably even change the back-up philosophy to take advantage of the huge levels of redundancy. There are a few downsides such as 'what happens during off-hours?', but the concept has potential.
On the off-chance that it hasn't already been invented (not likely), I hereby place the concept into the public domain. If so, then let's call it RAID-JeffyPooh (or RAID-JP). If it has already been invented, then please disregard (I didn't get the memo).
I'm no expert...
So, you virtualise your servers onto less hardware boxes, hasn't this just changed the ratio of 'hardware boxes performing server functions' to 'hardware boxes performing storage functions'. Do the storage boxes use significantly more power than they did before we all virualised our servers?
I'm not sure what my point is.
@ Hmmm... RAID-JP
If those desktops were running Linux they could use Lustre...
Is storage becoming IT's Chelsea tractor?
What was wrong with the chelsea tractor?
can't think of a comment. just preferred it over "hummer"
What's a good word for statistics masturbation? Vicent Flanders coined "flashturbation" for Flash overkill, but what can we coin for meaningless quotes of statistics?
There were no numbers or examples backing up anything in that article. If you start removing CPUs from an overall setup, of course the focus shifts to storage! But has the electricity usage increased or decreased from the previous setup??
Two different technologies
Virtualisation and centralised storage can both help to reduce power in different ways.
With virtualisation, you can reduce your processor count by making one piece of hardware work harder. The processor will of course use a few more watts, but overall by eliminating one or more machines you reduce the energy use.
With a SAN/NAS, you avoid having to purchase many servers with more storage than is initially required, or with the possibility to upgrade in the future (and hence beefier power supplies). Instead you concentrate all your storage and can reduce your drive count overall. Avoid having lots of little RAID arrays, and instead have one or more larger SANs or NAS. It's also much easier to manage capacity since upgrading the storage on a server (real or virtual) can be done in software.
Who cares if storage arrays start using proportionally more power when the number of servers and overall power consumption is going down?
Of course the downside of both approaches is that if something breaks, a good proportion of your services disappear suddenly :-)
Filthy yankee panderers
Wasn't the headline on this one about 'Chelsea Tractors' [UK] rather than 'Hummers' [potty mouthed gangsters and gubbernators] once upon a time? If you are going to pander that hard wouldn't a black and white costume and a handful of bamboo be in order?
Our memories last a little longer than you think Mr Ministry of Information! All go fetch my coat shall I? Cloakroom 101?
Sorry, can't agree!
Actually, every server consolidation exercise I've been involved in has seen the removal of many DAS units (usually SCSI) and their replacement with SAN arrays. Each of those DAS units, whilst also being underutilised storage, also had small, inefficient PSUs whereas the SAN arrays used larger and more efficient PSUs. In one datacenter I did the figures and I saw a drop in power costs where I could show the arrays were saving us 30% on power over the DAS units they replaced (plus also giving savings in admin, more efficiently utilised and therefore cheaper storage, easier replication, backups and DR, and savings in support costs). And that case was before thin provisioning which we now can use to allocate a large disk space in an array but start with only a small amount of the disk actually there, and then actually only add the disk as we get near the point where we need it (so less unused disks idly spinning away and chewing up current). In the survey mentioned, did they ask the same users that said they allocated more storage per server if their storage was better utilised and therefore likely to be less power-hungry after they went to consolidated virtualised servers?
Belt & Braces
Me, I'm an old geriatric engineer. I like redundancy, and would far rather see all those HDs in some form of RAID and all those REAL processors working together in some form of managed load sharing.
Oh, and as for inefficient PSUs. Just feed them with DC instead of AC and watch their efficiency soar
... as I think Google discovered.
Did I understand correctly?
That the VM disk storage will tend to be SAN-based as opposed to bundled in with the VM image?
This makes sense for replicated web serving VMs that all need to pull the same data for the same account from the same place regardless of who is up and who is not.
However, a lot of the attractiveness of VMs for non-datacenter types like me is their use as self-contained appliances. In this kind of application, bundling of the disk storage for the appliance's data into the VM image makes a lot more sense. The key considerations driving separate storage are whether the data needs to be shared with other applications or not, and if it has to be available when the appliance is turned off.
I think we're only in the launch phase of massive virtualization right now. Ultimately, the real benefits will turn out to be in the fields of convenience, backup/restore and disaster recovery; optimizing server utilization is just the salesman's foot in the door.
For those corporate and web applications architected for separate storage, I think we'll see the following developments fairly soon:
1. VMotion with synchronized dynamic DNS updates to preserve Internet visibility as well as application uptime.
2. DMotion (Data Motion) for SAN disaster recovery purposes so that the switching of VMs and conventional server applications to backup databases can proceed in an automatic and standards-governed way.
3. Virtualized SANs as an outgrowth of DMotion.
4. Increased incidence of motion sickness unless standards are developed.
- Breaking news: Google exec veep in terrifying SKY PLUNGE DRAMA
- Geek's Guide to Britain Kingston's aviation empire: From industry firsts to Airfix heroes
- Analysis Happy 2nd birthday, Windows 8 and Surface: Anatomy of a disaster
- Google CEO Larry Page gives Sundar Pichai keys to the kingdom
- Something for the Weekend, Sir? SKYPE has the HOTS for my NAKED WIFE