Describing the architectures. I am going to stop for today. I have a book signing at 10 and the WIT luncheon later today.
Unique fault tolerance approach
Talking about column store, sorting.
Now execution time increases by 1/3 with 4 nodes instead of doubling.
Divide rows across all other nodes.
Execution time will double even with a 100 node system. It’s like partition skew since one node will do double the work.
Mirror data to another node. One node can fail. Execution time will double.
Handling Hardware errors
Node failures with shared-nothing. Mirroring is not sufficient.
Explaining shuffling. This is familiar if you have worked with PDW/APS.
Joins, four cases depending on how tables are partitioned or replicated. Shuffling in some cases.
Explaining how a table is executed on all nodes. Same query plan runs on nodes.
Table replication. Tables are completely replicated on every node of the cluster.
Can lead to skew
There is no way to know where a row ends up. Hash key partitioning. Use hash key to determine where rows land.
Round-robin partitioning. Rows are distributed among storage units. All disks end up with the same number of rows.
Both shared and shared nothing use partitioned tables. This is the key to parallel processing. What does orthogonal mean?
Basis for scalable execution. Distribute the rows of the table across storage devices.
Network can limit scaling.
Remote storage, data is separated from compute. Local disks for tempdb.
Commodity servers and commodity networking. Storage and compute are co-located. Scales indefinitely.
APS and Azure DW are same software in different hardware form factors.
Two alternative scalable dw designs
Shared-nothing appliances such as APS
Shared-storage Oracle RAQ, Microsoft SQL DW
Scalable DW Fundamentals
Partitioned tables, but I think he means distributed tables??
Appliance is always best performing. Cloud is low cost.
Cloud – no CapEx low OpEx. Conception to insight in hours. Flexibility to scale.
Now, DW in the cloud.
DW Appliance, low compliance but high costs. OR roll your own with lots of complexity. That was what available in 2008.
What is driving datawarehousing. Cheap hardware and increasing amounts of data.
Time for David DeWitt. Our favorite keynote speaker!
Today is WIT lunch with Kelly Lockwood Primus. Tonmorrow is board Q&A. Speaker Idol round 2. Fill out your evals!
New website coming 2017
We connect, share, and learn. New Brand!
It’s a place that allows us to share. PASS brand was created in 1999
Denise McInerney. Our focus has shifted to the entire data platform. What does PASS mean to us.
Microsoft is our biggest sponsor. We have several Global Alliance Partner including SentryOne and Redgate!
To much to keep up with. Chapters, VCs, more.
Around 50% in North America.
We are one global community. We are growing around the world. SQL Saturday events continue to grow.
We include 87% of the countries of the world in our membership.
Grant Fritchey on the stage.
At least the guys wearing kilts are not also wearing high heels.
This is Kilt Day.
I am back at the blogging table for day 2.