SAP's support statement can be found on the following SAP Note: https://service.sap.com/sap/support/notes/800326
For many, this will have little meaning, but for some this will be a major breakthrough.
Let me state from the start that it's my true belief that for large enterprise customers, having a SAN based storage config is the best way to deploy SAP Systems, an in particular SAP HANA. My posts on this blog explain why i believe that, and it's a factor of performance, TCO, flexibility and operations aspects.
But the fact is that mainly in the service provider business, some organizations are looking to "white label" servers with direct attach storage, forming large pools of resources, as a cheaper infrastructure option in this very competitive "public cloud" world.
And in that case, you will want a couple of things:
- Performance, of course;
- But also redundancy;
- Scalability;
- and ease of management.
In the SAP world, so far, the only way to get the redundancy and scalability was using IBM's GPFS, which is a clustered file system providing many redundancy and scalability characteristics.
There are lots of merits and value in GPFS, but I think GPFS is the wrong tool for environments like databases (for example see my blog post on block vs file here). Also the feedback I'm getting from customers I meet using GPFS, say it's a nightmare to manage, implies unacceptable downtimes to scale, and implies as well operations costs that are not the most affordable in the market.
Well, ScaleIO is a technology that allows to aggregate the direct attach storage of many servers into a single "virtual SAN alike" block storage pool.
And as I've written on an earlier blog post, for database workloads, I believe that "block" access to disk is the right option. After all the databases read and write blocks, and the same happens with SAP HANA.
So, what ScaleIO allows is to:
- get redundancy into direct attach storage my ensuring a copy of each server's data on another server;
- get awesome performance by distributing the data of a filesystem attributed to a server accross all servers in the pool (all application servers are at the same time storage servers for their peers), which makes this solution to perform better the more nodes you have in the cluster (of course LAN network planning plays an important role here);
- scalability is natural in the design of ScaleIO, and the best is that you can add servers to the pool and re-balance volumes automatically behind the scenes. the same automatic re-balancing happens as well when you remove a server;
- and the best of it, is that many of these operations are done online.
Meanwhile, check out SAP Note 800326 for the official support statement from SAP for ScaleIO, and if you want to learn more about this technology, it's benefits and how it might fit on your datacenter strategy, drop me a message and I'll be happy to help.
One final note: if you are a fan of server based storage, you need to scale-out, you would like alternatives to GPFS, and would like to have more choice for this kind of setup either with IBM servers, or other server manufacturers, the best news of all, is that NOW you have!
Stay tuned as more new will come on this soon.