Samsung announces 3,000MB/s enterprise SSD

engadget.com

7 pointsRobAley3 comments

How much would this speed up say a MySQL DB? Would it be enough to eliminate the need to do some caching? Would it allow joins to be fast enough to not need to denormalize so soon?

My point is, does this buy out of the need (up to a certain scale) to use engineering time to solve scaling problems?

NOTE: I realize that you still want to use good design and plan for the appropriate amount of scale, I am just curious how much dev time this could potential save in terms of rewriting code for scale.

It can definitely save you some time (SSDs in general, not just this one,) but how much depends on the operations in question.

Example workflow - SQL Server, approx 1TB working set - putting it all on SSD (a RAID set of consumer level SSDs) basically doubled our operations per second. Modifying the algorithm a bit while still on the SSD gave 20x speed up.

2nd example - SQL Server, approx 10TB of RAID Prosumer SSD vs same amt of 15k SAS drives: SSD = 8x speedup.

3rd example - SSD as swap on linux operations - performance dropped by 10%. This mostly had to do with the SSD used and its role.

In other words - it's not a magic bullet but it can definitely buy you some breathing room.

"Algorithms are for people who don't know how to buy RAM."

Reply on news.ycombinator.com