This document discusses using AI to build a self-driving query optimizer called Unravel Sessions. It introduces Shivnath Babu and Adrian Popescu, who founded Unravel to focus on ease of use and manageability of data systems. The document outlines challenges with current approaches to application performance tuning. It then demonstrates how Unravel Sessions uses a probe algorithm and Gaussian process model to automatically recommend performance optimizations through iterative probes. Key lessons learned in building the Unravel Sessions architecture include using model ensembles, cheap probes when possible, full-stack monitoring data, and keeping the user in the loop.

1. Shivnath Babu,,Adrian,Popescu
Using&AI&to&Build&a&Self3
Driving&Query&Optimizer
#AI7SAIS

2. Meet$the$speakers
• Cofounder*and*CTO*at*Unravel,*Adjunct*
Professor*at*Duke*University
• Focusing*on*ease>of>use*and*manageability*
of*data>intensive*systems
• Recipient*of*US*National*Science*
Foundation*CAREER*Award,*three*IBM*
Faculty*Awards,*HP*Labs*Innovation*
Research*Award
• Data*engineer*at*Unravel
• PhD*from*EPFL,*Switzerland
• 8+*years*of*experience*in*performance*
monitoring*&*modeling*of*data*management*
systems
• Focusing*on*tuning*and*optimization*of*Big*
Data*apps
2#AI7SAIS
Shivnath Babu Adrian$Popescu

3. My#app#often#fails#with#Out#of#
Memory…
3
DATA#SCIENTIST
#AI7SAIS

4. How$can$I$make$it$more$reliable?
4
DATA$SCIENTIST
#AI7SAIS

5. My#app#is#too#slow…
5
DATA#ENGINEER
#AI7SAIS

6. I"need"to"make"it"faster…
6
DATA"ENGINEER
#AI7SAIS

7. My#app#is#missing#SLA…
7
DATA#PIPELINE#OWNER
#AI7SAIS

8. How$can$I$tune$my$app$to$guarantee$
SLAs?
8
DATA$PIPELINE$OWNER
#AI7SAIS

9. This%rogue%app%is%wasting%resources%
and%reducing%cluster%throughput
9
OPERATIONS%TEAMS
#AI7SAIS

10. Can$this$app$use$less$resources$
while$finishing$on$time?
10
OPERATIONS$TEAMS
#AI7SAIS

11. Current'approach:'Find'the'needle'in'
the'hay'STACKS
1"System
1"User
1"App
11
1. Find"the"app."Review"Spark/YARN"UI"to"find"the"app
2. Review"metrics"on"web"UI
3. Review"stages"associated"with"the"app
4. Identify"executors"associated"with"the"stages"
and"outliers
5. Deep"dive"into"“outlier”"stage
6. Identify"problematic"“executors”
7. Review"and"debug"container"logs
8. Rinse"&"repeat"across"other"executor/container"logs"
to"identify"the"problem
#AI7SAIS

12. Now$imagine$the$problem$at$scale
12
1$System
1$User
1 App
10x Systems
100x Users
1000x Apps
#AI7SAIS

13. Imagine(a(new(world
INPUTS
1. App(=(Spark(Query
2. Goal(=(Speedup
“I#need#to#make#this#app#faster”
#AI7SAIS

14. Imagine(a(new(world
14
TIME
APP(DURATION
In#blink#of#an#eye,#user#
gets#recommendations#to#
make#the#app#30%(faster
As#user#finishes#
checking#email,#she#
has#a#verified#run#
that#is#60%(faster
User#comes#back#from#
lunch.#A#verified#run#that#
is#90%(faster
90%(
faster!
#AI7SAIS

15. At#Unravel,#we#want#to#bring#
the#new$world$to#you#today
Introducing$Unravel$Sessions
15#AI7SAIS

16. Demo
16#AI7SAIS

17. Unravel(Sessions
17
App
Goal
SQL Program DAG Workload
Reliability
Efficiency
Speedup
SLA
TIME
Unravel(
Sessions(
will(quickly(
get(you(
there!
#AI7SAIS

18. Demo
18#AI7SAIS

19. How$Unravel$Sessions$work
AI
for$
Application$Performance$Management
19#AI7SAIS

20. Monitoring
Data
Historic-Data
&
Probe-Data
Recommendation
Algorithm
Cluster-Services-On=premises-and-Cloud
App,Goal
Orchestrator
Unravel-Sessions-Architecture
Xnext
Probe-Algorithm
#AI7SAIS 20

21. spark.driver.cores 2
spark.executor.cores
…
10
spark.sql.shuffle.partitions 300
spark.sql.autoBroadcastJoinThres
hold
20MB
…
SKEW('orders',@'o_custId') true
spark.catalog.cacheTable(“orders") true
…
We-represent-the-setting-as-
vector-X
X
PERFORMANCE
#AI7SAIS 21

22. • Goal:&Find&the&setting&of&
X&that&best&meets&the&goal
• Challenge:&Response&
surface&y&=&ƒ(X)$is&
unknown
X
PERFORMANCE
Given:&App&+&Goal
#AI7SAIS 22

23. Response'Surface'Methodology'
#AI7SAIS 23
Credit:(http://people.csail.mit.edu/hongzi/
Reinforcement'Learning

24. Model&the&response&surface&as
The&Gaussian&Process&model&captures&the&
uncertainty in&our&current&knowledge&of&the&
response&surface
)()()(ˆ XZXfXy t
+= !
!!
!
!!
)(Xf t
)(XZ
X
PERFORMANCE
Challenge:/Response/surface/
y/=/ƒ(X)/is/unknown
Here:
is&a&regression&model
is&the&residual&captured&as&a&
Gaussian/Process
#AI7SAIS 24

25. !
=
"#=
"=
)(
)(ˆ
*
*
)())(()(
Xyp
p
Xy dpppdfpXyXEIP
We#can#now#estimate#the#expected'improvement'EIP(X) from#
doing#a'probe'at#any#setting#X
Gaussian#Process#model#helps#estimate#EIP(X)
Improvement#at#any#
setting#X#over#the#best#
performance#seen#so#far
Probability#density#
function#(uncertainty#
estimate)
X
Opportunity
#AI7SAIS 25
PERFORMANCE

26. Get$initial$set$of$
monitoring$data$from$
history$or$via$
probes:$<X1,y1>,$
<X2,y2>,$…,$<Xn,yn>$
1
Select$next%probe
Xnext based$on$all$
history$and$probe$data$
available$so$far$to$
calculate$the$setting$
with$maximum%expected%%
improvement%EIP(X)
2
Bootstrap
Probe-Algorithm
Until-the--
stopping-
condition-
is-
reached
#AI7SAIS 26
PERFORMANCE
X

27. 4 6 8 10 12
02468
x1
y
4 6 8 10 12
02468
x1
y
4 6 8 10 12
02468
x1
y
4 6 8 10 12
02468
x1
y
X
Performance
!
EIP(X)
!
Xnext:"Do"next"
probe"here
This%approach%
balances%
Exploration%Vs.
Exploitation
!
Exploration
!
Exploitation
#AI7SAIS 27

28. Credit:(https://discovery.rsm.nl/articles/detail/1309how9to9balance9exploration9and9exploitation9in9multinational9enterprises
Data(Starved
&(High(Uncertainty
Data(Rich
&(Low(
Uncertainty
#AI7SAIS 28
App,Goal
Xnext
Probe(Algorithm

29. Monitoring
Data
Historic-Data
&
Probe-Data
Recommendation
Algorithm
Cluster-Services-On=premises-and-Cloud
App,Goal
Orchestrator
Xnext
Unravel0Sessions0Architecture
Probe-Algorithm
#AI7SAIS 29

30. What%is%the%
goal?
! Make%the%
app%faster
! Make%app%
reliable
! Make%app%
meet%SLA
! Make%app%
use%less%
resources
What%is%the%
app?
! SQL
! Program
! DAG
! Workload
Are%trained%
models%
available?
! Yes
! No
! Partially
Is%similar%
historic%data%
available?
! No
! A%little
! A%lot
How%soon%
user%needs%%
the%answer?
! Now
! Soon
! By%a%
deadline
#AI7SAIS 30
Probe%Algorithm:%Has%to%deal%with%a%
diverse%space%of%data%&%uncertainty

31. Dynamic(Selection
Xnext
Probe*Algorithm:*Use*
Model*Ensembles
Model(1 Model(2 … Model(N
#AI7SAIS 31

32. • Probe'Algorithm:'Use'model'ensembles
• Orchestrator:'Use'cheap'probes'when'possible
• Monitoring'data:'Use'full<stack'data'collection
• Recommendation'Algorithm:'Keep'user'in'the'loop
Lessons&Learned&Building&the&&&&&&&&
Unravel&Sessions&Architecture
#AI7SAIS 32

33. Monitoring
Data
Historic-Data
&
Probe-Data
Cluster-Services-On8premises-and-Cloud
Orchestrator
Xnext
Lessons)Learned:)Cheap)Probes
#AI7SAIS 33
Orchestrator-can-leverage-
cost8based-optimizers-like-
Catalyst-as-an-approximate,-
but-very-fast,-performance-
model-in-some-cases

34. Lessons&Learned:&Full&Stack
#AI7SAIS 34
Monitoring
Data
Historic-Data
&
Probe-Data

35. Monitoring
Data
Historic-Data
&
Probe-Data
Recommendation
Algorithm
Cluster-Services-On=premises-and-Cloud
App,Goal
Orchestrator
Xnext
Probe-Algorithm
#AI7SAIS 35
Lessons5Learned:5User5in5the5Loop

36. In#Summary
AI for$Application$Performance$Management
Sign$up$for$a$free$trial,$we$value$your$feedback!$
https://unraveldata.com/free;trial
Join$the$Unravel$team!
adrian,$shivnath [@unraveldata.com]
36#AI7SAIS