PASS Summit Keynote Live Blog – Day 2
20 October, 2013 (15:19) | SQL | By: Mark V
Behold my live blog for Dr. David DeWitt’s fabulous keynote.
<<Please click post title for Live Blog entries>>
Good morning from PASS Summit 2013. It’s Day 2, which means it’s time for the Dr. David Dewitt keynote! Thanks for following along.
Lights going dim. Here we go.
Douglas McDowell(PASS Executive VP) taking the stage.
Be sure to check out the Exhibitors. They make #Summit13 possible.
Over 75% of PASS funding comes from the annual Summit.
Second largest funding source is PASS Business Analytics Conference. 900 attendees. Raise $100k for PASS. Awesome.
PASS org broken into Portfolios, each with their own budget. Partitioning? (Couldn’t resist).
PASS has $1million in reserves. That is a great safety net for such an important organization.
PASS spent a little over $7.6 million in the past fiscal year. 30% of that spending outside the US. #GlobalInitiativesForTheWin
PASS IT has THREE people maintaining over 500 websites. Thanks for them for their work to keep us going.
PASS Bill Graziano taking the stage.
Recognizing Douglas McDowell who is migrating off the PASS Board. Thank you!
Recognizing Rob Farley who migrates off the board as well. Thanks, mate.
Recognizing Rushabh Mehta for contributions to PASS.
PASS Current VP Marketing, incoming PASS President Thomas LaRock taking the stage.
79 different countries tuned in to #PASSTV so far. That is fantastic!
Birds of a Feather Luncheon tomorrow. I am leading a table around BI. Can chew and chat.
SQLClinic a GREAT opportunity to get expert advice from Microsoft personnel.
Dr. David DeWitt taking the stage to discuss #Hekaton.
DeWitt: If you think my past keynotes have been hard to understand, take two Advil now.
Hekaton = OLTP Rocket Ship
Dr. DeWitt looking dapper in his “smart business casual attire.”
DeWitt: OLTP growing at a modest rate. We see more of these workloads fitting in memory.
CPUs not getting faster. That well is dry.
Getting 100X improvement impossible with today’s software. Would have to reduce number of instructions by 99%.
DeWitt will explain why pinning tables in memory will not do the trick to improve performance dramatically.
I will try to keep up. #TallOrder
Dr. DeWitt explaining how latches can be a very contentious aspect of the Buffer Pool. Can be a big perf hit.
Dr. DeWitt now going into Serial Schedules while discussing Concurrency. #LearningStuff
Jim Gray, DeWitt’s late mentor, created the idea of 2 Phase Locking for Concurrency.
Two Rules of 2 Phase locking: 1. Query must acquire appropriate Lock. 2. Query cannot get a new once it releases a Lock.
The number of brilliant people who have made my career in #sqlserver possible is staggering. #Wow
#Hekaton uses Lock-Free data structures. No latches. Timestamps + optimistic concurrency control. Compilation into DLL.
Huge thanks to the DBAs who go deep into Internals so that BI folk like me don’t have to. #Respect
How do you use #Hekaton? Create a MEMORY_OPTIMIZED table. VERY simple to do.
EVERY #Hekaton table must have a Primary Key: either HASH or RANGE.
#Hekaton seeing 5X to 30X performance boost.
Diving into details of Lock-Free data structures… Oh boy.
DeWitt: “I can barely understand Lock-Free data structures.” #Yikes
Lock-Free data structures are the only way to get rid of the latches, spin-locks, etc that get in the way of perf
Dr. DeWitt’s animations for showing Lock-Free data structures is AWESOME.
Lock-Free structures allow any number of reader and updater threads to do their work with NO performance hit.
#Hekaton took 5 years for a VERY good reason. Hard stuff right there.
#Hekaton uses Multiversioning for Concurrency. Changes create new rows. #TypeIISCDForTheWin?
DeWitt going into Transaction Phases in #Hekaton: Normal Processing, Validation, Post-Processing
Reliance on Timestamps allows each Transaction to get the proper version of any row to read.
Since each Transaction gets a unique number, it can maintain proper order of tran commits without locking.
Presenters please not that Dr. DeWitt has not read to us once. He is talking, not reading.
#Hekaton End Timestamps are always unique. Very important concept there.
#Hekaton uses Timestamps both on the Rows AND on the Transactions.
Dear Dr. DeWitt, passion and enthusiasm are a huge asset for a presenter. Keep letting it shine!
#Hekaton has a garbage collector to remove the expired rows once they are not usable: based on Timestamps.
Wrapping up: Updates create new versions. Trans use Timestamps and versions for concurency.
Dr. DeWitt digging into Isolation Levels as they relate to the Validation phase of #Hekaton transactions.
Note: #Hekaton Phantom Avoidance does not involve Buffy the Vampire Slayer.
DeWitt digging into Post Processing phase. I will try to sum it up.
Normal Processing: transactions go through processing without locks (for the most part).
Validation: Determines if the transaction can be safely committed.
Post-Processing: wrapping it all up
A big key for #Hekaton: Stored Procedures rule.
Physical query execution plans likely to be very different in #Hekaton than “classic” sql server.
dll created in #Hekaton is targeted for that query. VERY lean.
I hate to bail on this early. But I need to get prepped for my presentation at the other end of building. :(
Thanks for following! And huge thanks to Dr. DeWitt for an awesome presentation as always.