Scale-out or Scale-up – that is the question! I’m preparing for my SAPPHIRE/ASUG session (A4526: Scale-up Architecture Makes Deploying SAP HANA Simple) and I’m trying to get a sense for what new HANA implementations are planning when it comes to hardware infrastructure.
With a fear of showing my age, I want to draw a comparison to a time when scale-out data centers were so common that the term scale-out wasn’t used because it was considered normal. Last night’s episode of Silicon Valley (the data center scene for you SV fans) made me think back to my early days as an MVS systems programmer – we outgrew our data center and had to build a new one with several times the square footage as the original. While it seemed like a great idea at the time, Moores Law and improvements in software reduced the required hardware footprint by so much that we ended up with enough unoccupied raised-floor space to accommodate a Coldplay concert!
Throughout the years, multi-node solutions have been replaced with single node equivalents – back in the day, large SAP deployments were based on multi-node relational databases because of capacity requirements. We wouldn’t even consider doing this today (with a few exceptions).
By now, I’m sure you can see that it is my opinion that only the absolute biggest BW systems (greater than 16TB of HANA memory) require the scale-out model. I’m surprised when I read forums and blogs where customers describe their up-coming scale-out deployments. I believe that the scale-out approach adds a significant amount of complexity while providing only minor benefits. With that said, I’m sure that some of you disagree with my opinion and have thoughtful reasons as to why you prefer a scale-out approach.
Please comment as you see fit and stop by my session on Thursday at 2PM so we can compare notes.