StorageBod I see quite a few posts about this storage or that storage ... how it is going to change everything or has changed everything. And yet, I see little real evidence that storage usage is really changing for many. So why is this?
Let’s take on some of the received wisdom that seems to be percolating.
Object storage can displace block and file?
It depends. Replacing block with object is somewhat hard. You can’t really get the performance out of it – you will struggle with the APIs especially to drive performance for random operations and partial updates.
Replacing file with object is somewhat easier. Most unstructured data could happily be stored as object and it is. It’s an object called a file. I wonder how many applications even using S3 APIs treat Object Storage as anything other than a file-store, how many use some of the extended metadata capabilities?
In many organisations what we want is cheaper block and file. If we can fake this by putting a gateway device in front of object storage, that’s what we will do. The object vendors have woken up to this and that is what they are doing.
But if a vendor can do "native" file with some of the availability advantages of a well-written erasure coding scheme at a compelling price point, we won’t care.
And when I can boot from object storage...call me.
All new developers are object storage aficionados?
I’m afraid from my limited sample size; I find this is rarely the case. Most seem to want to interact with file-systems or databases for their persistence layer. Now the nature of the databases that they want interact with is changing with more becoming comfortable with NoSQL databases.
Most applications just don’t produce enough data to warrant any kind of persistence layer that requires Object or even any kind of persistence layer at all.
Developers rarely care about what their storage is, they just want it to be there and work according to their needs.
Technology X will replace technology Y
Only if Technology Y does not continue to develop and only if Technology X has a really good economic advantage. I do see a time when NAND could replace rotational rust for all primary storage but for secondary and tertiary storage? We might still be a way off.
It also turns out that many people have a really over-inflated idea about how many IOPs their application need; there appears to be a real machismo about claiming that you need 1,000s of IOPS… when our monitoring shows that someone could write with a quill pen and still fulfil the requirement. Latency does turn out to be important – when you do your 10 IOPS, you want it to be quick.
Storage is either free or really cheap?
An individual gigabyte is basically free; a thousand of these is pretty cheap but a billion gigabytes is starting to get a little pricey.
A terabyte is not a lot of storage?
In my real life, I get to see a lot of people who request a terabyte of storage for a server because hell, even their laptop has this amount of storage. But for many servers, a terabyte is a huge amount of storage ... many applications just don’t have this level of requirement for persistent data. A terabyte is still a really large database for many applications – unless the application developers haven’t bothered to write a clean-up process.
Software-defined is cheaper?
Buy a calculator and factor in your true costs. Work out what compromises you might have to make and then work out what that is worth to you.
Google/Amazon do it, so we can too?
You could but is it really your core business? Don’t try to compete with the web-scale companies unless you are one ... focus on providing your business with the service it requires.
Storage administration is dead?
It changed, you should change too but there is still a role for people who want to manage the persistent data-layer in all its forms. It’s no longer storage … it’s persistence. ®