Questions and Answers, part one: Red Hat Storage Server 3, Ceph, and Gluster

Please note: Red Hat Storage Server has a new name: Red Hat Gluster Storage. Learn more about it, and Red Hat Ceph Storage, here: http://red.ht/1Hw7gYb

You may recall we recently launched Red Hat Storage Server 3 (learn more about that here). Well, we had a lot of questions arise during our keynote that we weren't able to share at the time due to the live nature of the broadcast. Well, we've collected all the questions we received and compiled a series of blog posts around them. We'll be sharing them over the next few days...starting with this post.

Competitor Comparison

What proof points do you have for the storage TCO compared to either legacy methods or your competitors?

IDC published a white paper that details this information. Check out The Economics of Software-based Storage report.

How is RHS Ceph on commodity hardware positioned for advantage against big storage vendors like NetApp, EMC, HDS, etc...

Red Hat and open source gives customers freedom of choice on hardware which helps them to drive down costs. The scale-out architectures of Red Hat Storage Server and Red Hat's InkTank Ceph Enterprise are better suited for massive data growth without having to invest upfront. In addition, as we run on industry-standard hardware and combine Red Hat Enterprise Linux with GlusterFS as the underlying storage OS the storage nodes can also be used to run some infrastructure applications which helps to reduce datacenter footprint and costs.

How does this differ from Netapp & EMC?  What advantages does it have over these big storage players?

NetApp has no real scale-out storage solution. EMC does have Isilon for scale-out file storage which is a proprietary appliance with similar features, but at significantly higher costs. Red Hat Storage also provides converged (computing and storage) capabilities, whereas NetApp and EMC do not.

Can you provide Red Hat's definition for Software Defined Storage and how the capacity and security mechanisms in SDS are improved & differentiated compared to traditional storage?

Software-defined storage decouples the storage hardware and the control layer. Also the common SDS regime is to use standard hardware and develop advanced features rather in the software layer. RHS uses RHEL as the underlying storage OS which provides military grade security features. Using a mainstream OS like RHEL also means that overall more customers are using it and when security issues are discovered that they can receive fixes immediately. Just as a recent example it took RH a couple of hours to fix a very wide-spread and dangerous security issue in the bash shell (shellshock) whereas there are quite a few UNIX, linux or freebsd based deployments or appliances which still don't have a fix for this issue as of today.

How do you measure advantage your system provides?

We can measure three factors: Costs, Scalability and Performance. Our costs are on average up to 60% lower than our competition, we can now scale up to >30 PB in a single pool and linearly increase throughput by just adding additional storage nodes.

What are the advantages of RHSSv3 compared to ZFS solutions like Nexenta?

ZFS is a great file system but it's running on a single-node and therefore can't really scale-out but rather uses the scale-up approach (adding CPU, cache, SSD). This is fine for small to medium environments, but brings the same limitations as proprietary legacy storage appliances. They don't scale.

How is software-defined storage different from storage hypervisors or storage virtualization?

Red Hat Storage Server and Red Hat's Inktank Ceph Enterprise use virtualization approaches as well but they go beyond that capability and provide many more features. Also the most commonly used storage virtualization technologies are block-based and provide just larger virtual block pools which usually require more expensive and complex fibre-channel based storage. RHS pools and virtualizes filesystems based on commodity storage servers where there is no need for a shared storage system or fibre-channel, but rather uses the disks which are in each of the storage servers. We use an algorithmic approach for virtualization which tends to scale better than classic storage virtualization technologies which have to go through a controller appliance or metadata server.