Original URL: http://www.theregister.co.uk/2013/10/07/storage_cloud/

Microsoft's Azure will bring tiers to your eyes

The joy of automated backup

By Aaron Milne

Posted in Cloud, 7th October 2013 09:30 GMT

For those of us who have been in IT for a while the term hierarchical storage management (HSM) has become more than a little old fashioned.

HSM was coined back in the bad old mainframe days when a high-capacity 300MB disk cost tens of thousands of dollars. At that time it made sense to have a primary array of disks backed up by a much slower, but far cheaper, tape library.

There are certainly some instances today where running an HSM system would appear to still make financial sense. Organisations such as the CSIRO in Australia, for example, are still heavily dependent on HSM.

Now cloud providers are coming of age and 200TB of consumer-grade disk can be had for as little as £20,000 (for the disks alone).

Heavenly pair

If you had asked me six months ago if running a traditional HSM system was a good idea my answer would have been a resounding “sometimes”. My answer today, however, would be "absolutely not."

Although Microsoft’s new Azure "on-premises" will one day soon allow for the creation of Azure and Hyper-V-compatible hosted clouds to be stood up anywhere in the world, that day has not yet arrived.

What we do have is Hyper-V and its near Azure-like ability to create local clouds. Combine this with Windows Server 2012 R2 and what you have is a pairing made in heaven.

Given then that we can set up our own private cloud to rival a Microsoft-hosted instance of Azure, does the question change from "should we run our own private cloud option?" to "why shouldn’t we"?

Let’s have a look.

Hot and cold

The first thing we should really do is kill off the concept, and the inherent baggage that the term brings with it, of HSM. It has had its day and deserves to fade into the sunset with a planned obsolescence.

In its place I’d like to suggest we replace it with a term that we should starting to be more than a little familiar with: automated storage tiering (AST).

This is really just a way to describe the different types of storage we are running in our data centres. Anyone familiar with virtualised data centres should have a working familiarity with the concept.

The usual breakdown of AST is as follows:

Hot PCIE-based flash drives serial attached SCSI solid state drives (SSDs);

Cold Traditional spinning rust hard disk drives (HDDs).

Given the increased availability and dramatic drop in price of flash-based disks and SSDs, most of us will have a few of these in play. AST is really just automating the movement of heavily accessed files between our hot SSD storage and our colder HDD storage.

This sounds great in theory but in practice it hasn’t been all that easy to set up or manage. Some vendors offer appliances that can do it for us but these can cause lock-in and they all come with their own costs and caveats. So really they are not an optimal solution.

Microsoft has taken everything it has learned and thrown it into an interface

Storage Spaces in Server 2012 R2, on the other hand, are the bees' knees, the ants' pants or any other insect that we can assign clothing to.

Microsoft has taken everything it has learned in making its cloudy Azure service work nicely and thrown it into an interface and back end. It will take your JBODs and your physically attached storage devices and allow you to pool them.

It will also allow you to create automated storage tiers in three easy steps. Although it is untested (and probably more than a little stupid), as well as definitely unsupported, you should be able to create point-to-point virtual private networks and then attach your remotely hosted device on Azure Server 2012 storage as an iSCSI storage space.

If you have local data centre silos, and the right links you could plausibly gain performance that is similar to that of your locally present HSM.

Pay the price

So you have set up your test lab, you have proofed your concept and now you are ready to roll out your first test cases and servers. We have even had the elusive signoff from the CIO for some Azure hosting.

Once we get to this point we would normally get our local Microsoft rep in to have a little chat about pricing. If you are a large customer of Microsoft’s you should be able to wangle a decent discount. Otherwise you are likely to pay the sticker price listed on the Azure website.

You might want to make sure you are sitting down for the next part if you have a weak constitution, because this might hurt.

Say we need 200TB of storage because what we really want is to replace that pesky HSM array and therefore we are migrating everything in it to the cloud.

To do this locally, we could probably get away with paying at most £70,000 for the disks, enclosures, servers and the associated electricity costs over three years.

With the prices quoted under Pay As You Go on the Azure website, though, you would be looking at somewhere in the order of four times as much.

If you pay 12 months in advance you can reduce this by up to 32 per cent. Even with a discount of a third we would still be paying about £130,000 more to host our data in Microsoft's cloud than we could reasonably expect to pay to run our own local cloud.

By sending our data rocketing into the cloud we could also be opening ourselves up to a world of pain with regards to data sovereignty.

Microsoft is rushing to build up and roll out locally hosted Azure options in countries around the world, but these are not in place everywhere yet. There is no guarantee that once they are you will be able to migrate your services easily or at a reasonable cost into your locally hosted cloud.

You might think this doesn’t matter. It is certainly possible to encrypt your virtual hard disks and to further encrypt the data inside them, but are you absolutely certain that the programs and encryption you will be using haven’t been compromised?

The cloud continues to be one of the least secure options available from a data sovereignty point of view.

Fatal attraction

So given that we might not be able to utilise a remotely hosted cloud offering because it is so expensive at scale, is there a reasonable level where it remains affordable? Also, does it have any possible use at all given the data sovereignty concerns?

Actually, it does. In talking up certain aspects of how Azure handles your data at the data centre level, Microsoft has all but beaten us over the head with what remotely hosted Azure is really good for: providing an encrypted backup storage service.

For 50TB of storage on the Pay As You Go plan you will pay $3,333 per month (Microsoft’s website quotes prices in US dollars but bills customers in various countries in local currency).

Now that might seem like a lot for many businesses, considering that it clocks in at $119,988 for 36 months. But given that backups are a cost of doing business and a reliable backup service can often cost about this much anyway, it is definitely worth a look.

Small beginnings

It becomes an even better option once you realise that Microsoft has baked the necessary hooks to make it happen into the Windows Server 2012 R2 Backup and Restore Center.

So while we can’t store our clients’ data on Azure, we might just be able to store our own.

Azure as a cloud service is a fantastic option for development and for low-volume storage but it doesn’t really scale affordably yet.

Microsoft is working feverishly to get Azure hosting up just about everywhere and once this – or the option for wholly on-premise Azure – kicks in you should grab hold of it with both hands.

For now, use it as a wonderfully reliable online backup service and keep watching the skies. ®