The next unified storage

The last week of October I’ve been to Storage Networking World Europe, a SNIA event in Frankfurt. I attended a lot of presentations and met interesting people.

I tried to collect opinions and ideas and now I write this post to share my point of view.

A lot of presentations and talks started from the same observed trend: data storage is exploding (not a great piece of news, indeed). Not structured but unstructured data is becoming the big pain for enterprises, we are seeing a huge growth of everything stored in our arrays and the IT managers are pretty conscious that they need to find a solution ASAP. Data explosion is not all about new data but also about the fact that you can’t delete or archive properly and you have no efficient way to manage them correctly: Storage is hostage of users, bad processes and procedures and, many times, limited budgets.

Contrary to some predictions, we will not see the death of block storage. Block storage will remain the primary option for DBs and critical applications thanks to its low latencies and great performance but I’m pretty sure that file based service protocols (like CIFS or NFS) will gain the king role in the future datacenters because of their ease of use with average to good performances. Block+Files storage arrays (the so called unified storage) aren’t enough to respond to future needs: the picture isn’t complete and we need more.

The huge amount of data to store needs cloud storage to be managed (private and/or public), We need to think no more in terms of blocks or files but in a smarter way with a new atomic measurement unit: the object (data+metadata). Indeed, this way to manage data isn’t new and it is in use since decades in many companies in different forms (do you remember CAS? content addressable storage) but with proprietary, dedicated and expensive systems: now things are changing.

The cloud storage is often a repository of objects accessible via APIs, like for CAS in the past. Vendors and associations are working very hard to impose one or, at least, few standards. I could add that we haven’t a standard yet but many APIs are pretty similar and I think that we will see a big shrink in the number of player in the next few years.

The future is not clear, but the wide adoption of cloud storage will help very much: the winners will impose the standards and othe vendors will use it to do better things.

I’m sure you have read many times about the end of RAID (in terms of its capability to protect data). Some analysts talk about 2020, some others are looking to a more near term but no one has doubts about its end of life (in terms of security of your data). Of course, I’m not sure that RAID will die, probably we will see two scenarios: one for structured data and one for unstructured data.

In the structured data domain we will se an evolution of actual raid mechanism (perhaps, a triple parity RAID or a dual Mirror?), basically for two reasons: traditionally structured data isn’t so big and you need very predictable performance and latencies to grant the right service level.

In the unstructured data domain things will be quite different, you look more for throughput than IOPS or latencies. New data protection mechanisms are based on next generation algorithms and they are designed with the cloud in mind: multi node/site data dispersion, mixed level of reliability, tiering, great automations and so on.

We can find one of the most sparkling segments of the whole IT industry in the scale-out storage. The big vendors already have (often acquired) or are looking for this kind of solutions. Some of them are Block based (i.e.: Equallogic from Dell) others are file based (SONAS from IBM or Ibrix from HP) and there are Object based too (see the latest evolutions from Caringo, NetApp/Bycast, etc.). All the vendors want a piece of this, relatively, new market.

Scale-up monolithic storage, also in the latest clustered implementations, have a limited scalability and they cost too much to be justified for files and object storage.

Scale-out storage systems have a pretty unlimited scalability, embedded replicas, ease of management and they are based on commodity hardware.

Scale out storage is the base of public clouds but also they are becoming the base of private clouds. First to adopt these storages were the big internet/cloud providers but things are changing rapidly and enterprises are catching up curiosity for these new technologies and companies.

If it is true that unified storage (block+files) on converged media (iSCSI, FC, NFS, CIFS are all deployed Ethernet) is the present what can we expect for the next future? A more unified and converged storage architecture.

The media will remain Ethernet (companies are still investing in its evolution) the protocols will be FCoE/iSCSI, pNFS and an Object storage (on HTTP). I guess that modular scale-out storage will be the standard for all vendors, not tomorrow of course but I think this is the way and we aren’t so far away…