Suspend disbelief for a moment and ponder the following questions: What would your business look like if we made it economical to run your storage infrastructure at speeds closer to that of memory than that of disk storage? How fast would your applications run? What business models would it open up? How much competitive advantage can you gain?
Modern hosting and service providers often face the challenge of managing the cost efficiencies of their platform. A common solution is to deploy a multi-tenant or multi-instance architecture in which many customers share the same hardware. The reuse of hardware over many clients drives down costs and also reduces the required ongoing administration.
Multi-tenant deployments host many clients in the same instance of software while segregating the client data through configuration. Multi-instance designs are similar but run one instance of software per client. To the storage tier, both approaches require many sub-ecosystems to run simultaneously in a shared space and will cause similar access challenges:
- Unpredictability of usage
- Height of individual usage spikes
- Scale versus storage performance (more clients translates into a more random and parallel workload)
SSD’s are like cordless phones and DVD’s. They made an improvement on an existing technology but didn’t revolutionize its use. In technology there is a difference between the concept of modernizing and revolutionizing. Modernizing is finding a way to do the same thing, just a little bit faster or little bit easier. Revolutionizing is either eliminating or vastly changing how something is done.
Flash is about latency and IOPs so why would it be good for Data Warehousing or Business Intelligence?
Excellent question. Yes, the typical marketing and wow-factor stats around flash are based on latencies and IOPs (Input Output Per Second). Data warehousing (DW) and Business Intelligence (BI) is normally a throughput game, so what gives?
Transactional workloads are commonly defined as being small atomic pieces of work. This is in contrast to decision support, Data Warehouse, Business Intelligence or otherwise labeled reporting systems that require fewer, larger, more sequential workloads. Updates, inserts, deletes and even small result-set selects are all included in OLTP, transactional efforts.
Violin's own Ashminder Ubhi, an Oracle expert, recently tested the OLTP (online transaction processing) storage performance of Violin Systems arrays vs. Exadata and SSD-based solutions using the popular SLOB benchmarking tool. Check out Ubhi's blog post on OLTP performance for the full story.
Ubhi's benchmarks highlight how a purpose-built flash memory array can offer benefits over products which are not designed to get the performance benefits of NAND flash technology – especially for an OLTP workload.
Back to school day.
Over at Stanford they have a speaker series that has been going on for the last few... decades, called the Stanford University Department of Electrical Engineering Computer Systems Colloquium, known to many simply as EE380. The list of past speakers is as they say, long and distinguished, and includes such industry lightweights as Joy, Lamport, Colwell, Bechtolsheim, Gray, Metcalfe, Gelsinger, Hennessy, Patterson, Brin & Page, Diffie, Mashy, Wolfram, Cerf and Kay. For a mere mortal being invited to give an EE380 talk can be an intimidating experience, which they try to make easier by telling you that there will probably be no more than 50 people in the room, thankfully they didn't mention that 10,000 people will watch the web cast online until after my talk was over. Yes, the other day Bennett and Rowett were added to the list of "past EE380 speakers."
Topics: dedup, Flash Array, groomer, MLC, solid state drive, Computer Data Storage, flash memory, flash storage, garbage collection, grooming, IOPS, jon bennett, memory array, memory arrays, PCIe, server, SLC, SSD, Systems Design, Storage Array
Start From Flash
There is discussion in the industry as to the best method to deploy solid state storage, in other words whether it is better to build flash arrays from SSD's, PCIe Cards, or the flash chips themselves. The key to system design is selecting the right building blocks and integrating them in such a way that the overall system is optimized - while minimizing tradeoffs. Violin strongly believes that you need to start from the flash layer (chips), and build an array as a coherent, tightly integrated whole to extract the extraordinary value offered by these amazingly high performance flash die.
Why did we take disk drives out of servers? Now we rely on traditional disk array storage and surround it with data center compute. The industry doubles processing power every 12 to 18 months and Gigabit Ethernet and Infiniband now provide 10 – 40Gig bandwidth with very low latencies. So we have lots of compute and we are swimming in bandwidth but the storage array hasn’t evolved at even 1/100 the pace? The lack of IO is killing the applications.
Now in 2012 it seems clear that solid state storage will be the solution to balance the network-compute-storage triangle and provide the IO necessary to not only virtualize the easy applications but now tackle the hard (latency-sensitive) IO bound applications that have been fine tuned to run on dedicated servers.
George Crump of Storage Switzerland shot another chalk talk video, the subject of this one is sharing solid state storage. George added it to his article from a month ago where he shared his thoughts on Violin's certification with IBM SAN Volume Controller (SVC).