Changing the game with cloud dw


Published on

Snowflake presentation by Kent Graziano at Houston Hadoop & Spark Meetup on 11/29/2016

Published in: Technology
1 Like
  • Be the first to comment

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Changing the game with cloud dw

  1. 1. 1 Y O U R   D A T A ,   N O   L I M I T S Kent  Graziano   Senior  Technical  Evangelist Snowflake  Computing Changing  the  Game  with   Cloud  Data  Warehousing @KentGraziano  
  2. 2. 2 My  Bio •Senior  Technical  Evangelist,  Snowflake  Computing •Oracle  ACE  Director  (DW/BI) •OakTable •Blogger  – The  Data  Warrior •Certified  Data  Vault  Master  and  DV  2.0  Practitioner •Former  Member:  Boulder  BI  Brain  Trust  (#BBBT) •Member:  DAMA  Houston  &  DAMA  International •Data  Architecture  and  Data  Warehouse  Specialist •30+  years  in  IT •25+  years  of  Oracle-­related  work •20+  years  of  data  warehousing  experience •Author  &  Co-­Author  of  a  bunch  of  books  (Amazon) •Past-­President  of    ODTUG  and  Rocky  Mountain  Oracle   User  Group  
  3. 3. 3 Agenda •Data  Challenges •What  is  a  Cloud  Data  Warehouse? •What  can  a  Cloud  DW  do  for  me? •Cool  Features  of  Snowflake •Other  Cloud  DW  – Redshift,  Azure,  BigQuery •Real  Metrics
  4. 4. Data  challenges  today
  5. 5. 5 Scenarios  with  affinity  for  cloud Gartner  2016   Predictions: By  2018,  six   billion  connected   things  will  be   requesting   support. Connecting  applications,  devices,  and   “things” Reaching  employees,  business  partners,   and  consumers Anytime,  anywhere  mobility On  demand,  unlimited  scale Understanding  behavior;;  generating,   retaining,  and  analyzing  data
  6. 6. 6 40 Zettabytes by 2020 Web ERP3rd party  apps Enterprise  apps IoTMobile
  7. 7. 7 It’s not the data itself it’s  how  you  take  full  advantage  of  the  insight  it  provides Web ERP3rd party  apps Enterprise  apps IoTMobile
  8. 8. 8 All  possible data All  possible actions Most  firms  don’t  consistently  turn  data  into   action 73% 29% of  firms   aspire  to  be   data-­driven. of  firms  are   good  at  turning   data  into   action. Source:  Forrester
  9. 9. 9 New  possibilities  with  the  cloud •More  &  more  data  “born  in  the  cloud” •Natural  integration  point  for  data •Capacity  on  demand •Low-­cost,  scalable  storage •Compute  nodes
  10. 10. 10 Cloud  characteristics  &  attributes DYNAMIC EASY FLEXIBLE SECURE Scalable Elastic Adaptive Lower  cost Faster   implementation Supports  many   scenarios   Trust  by   design
  11. 11. 11 The  evolution  of  data  platforms Data  warehouse   &  platform   software Vertica,   Greenplum,   Paraccel,  Hadoop, Redshift Data   warehouse   appliance Teradata 1990s 2000s 2010s Cloud-­native   Data   Warehouse Snowflake 1980s Relational   database Oracle,  DB2, SQL  Server
  12. 12. 12 What  is  a  Cloud-­Native  DW? •DW-­ Data  Warehouse •Relational  database •Uses  standard  SQL •Optimized  for  fast  loads  and  analytic  queries •aaS  – As  a  Service •Like  SaaS  (e.g. •No  infrastructure  set  up •Minimal  to  no  administration •Managed  for  you  by  the  vendor •Pay  as  you  go,  for  what  you  use
  13. 13. 13 Goals  of  Cloud  DW •Make  your  life  easier •So  you  can  load  and  use  your  data  faster •Support  business •Make  data  accessible  to  more  people •Reduce  time  to  insights •Handle  big  data  too! •Schema-­less  ingestion
  14. 14. 14 Common  customer  scenarios Data  warehouse  for   SaaS  offerings Use  Cloud  DW as  back-­ end  data  warehouse   supporting  data-­driven   SaaS  products noSQL  replacement Replace  use  of  noSQL   system  (e.g.  Hadoop)  for   transformation  and  SQL   analytics  of  multi-­ structured  data   Data  warehouse   modernization Consolidate  legacy   datamarts  and  support   new  projects
  15. 15. 15 Over 250 customers demonstrate what’s possible Up  to  200x  faster  reports  that  enable  analysts  to  make   decisions  in  minutes  rather  than  days Load  and  update  data  in  near  real  time  by  replacing  legacy   data  warehouse  +  Hadoop  clusters Developing  new  applications  that  provide  secure  (HIPPA)  access   to  analytics  to  11,000+  pharmacies
  16. 16. 16 Introducing  Snowflake
  17. 17. 17 About  Snowflake   Experienced,   accomplished leadership  team 2012   Founded  by   industry  veterans   with  over  120   database  patents Vision:   A  world  with   no  limits  on  data First  data warehouse built  for  the   cloud Over  230 enterprise   customers  in one  year since  GA
  18. 18. 18 The  1st Data  Warehouse  Built  for  the  Cloud Data  Warehousing… • SQL  relational  database • Optimized  storage  &  processing • Standard  connectivity  – BI,  ETL,  … •Existing  SQL  skills  and  tools •“Load  and  go”  ease  of  use •Cloud-­based  elasticity  to  fit  any  scale Data   scientists SQL   users  &   tools …for  Everyone
  19. 19. 19 Concurrency Simplicity Fully  managed  with  a   pay-­as-­you-­go  model.   Works  on  any  data Multiple  groups  access   data  simultaneously   with  no  performance  degradation Multi  petabyte-­scale,  up  to  200x  faster   performance and  1/10th  the  cost 200x The  Snowflake  difference Performance
  20. 20. 20 The  Data  Warrior’s Top  10  Cool  Things About  Snowflake (A  Data  Geeks  Guide  to  DWaaS)
  21. 21. 21 #10  – Persistent  Result  Sets •No  setup •In  Query  History •By  Query  ID •24  Hours •No  re-­execution •No  Cost  for  Compute
  22. 22. 22 #9  Connect  with  JDBC  &  ODBC Data  Sources Custom  &  Packaged   Applications ODBC WEB UIJDBC Interfaces Java >_ Scripting Reporting  &   Analytics Data  Modeling,   Management  &   Transformation SDDM SPARK  too! JDBC  drivers  available  via Python  driver  available  via  PyPI
  23. 23. 23 #8  -­ UNDROP UNDROP  TABLE  <table  name> UNDROP  SCHEMA  <schema  name> UNDROP  DATABASE  <db  name> Part  of  Time  Travel  feature:  AWESOME!
  24. 24. 24 #7  Fast  Clone  (Zero-­Copy) •Instant  copy  of  table,  schema,  or   database: CREATE OR  REPLACE   TABLE MyTable_V2 CLONE MyTable • With  Time  Travel: CREATE SCHEMA mytestschema_clone_restore CLONE testschema BEFORE (TIMESTAMP => TO_TIMESTAMP(40*365*86400));;
  25. 25. 25 #6  – JSON  Support  with  SQL Apple 101.12 250 FIH-­2316 Pear 56.22 202 IHO-­6912 Orange 98.21 600 WHQ-­6090 { "firstName": "John", "lastName": "Smith", "height_cm": 167.64, "address": { "streetAddress": "21 2nd Street", "city": "New York", "state": "NY", "postalCode": "10021-3100" }, "phoneNumbers": [ { "type": "home", "number": "212 555-1234" }, { "type": "office", "number": "646 555-4567" } ] } Structured data (e.g. CSV) Semi-structured data (e.g. JSON, Avro, XML) • Optimized storage • Flexible schema - Native • Relational processing select  v:lastName::string as last_name from  json_demo;;
  26. 26. 26 #5  – Standard  SQL  w/Analytic  Functions Complete SQL database • Data  definition  language  (DDLs) • Query  (SELECT) • Updates,  inserts  and  deletes  (DML) • Role  based  security • Multi-­statement  transactions select  Nation,  Customer,  Total from  (select   n.n_name  Nation, c.c_name  Customer, sum(o.o_totalprice)  Total, rank()  over  (partition by  n.n_name order by  sum(o.o_totalprice)  desc) customer_rank from  orders  o, customer  c, nation  n where  o.o_custkey  =  c.c_custkey and c.c_nationkey  =  n.n_nationkey group  by  1,  2) where  customer_rank  <=  3 order  by  1,  customer_rank
  27. 27. 27 Snowflake’s multi-cluster, shared data architecture Centralized  storage Instant,  automatic  scalability  &  elasticity Service Compute Storage #4  – Separation  of  Storage  &  Compute
  28. 28. 28 #3  – Support  Multiple  Workloads Scale  processing  horsepower  up  and down  on-­ the-­fly,  with  zero downtime  or  disruption Multi-­cluster  “virtual  warehouse”  architecture  scales   concurrent  users  &  workloads  without  contention Run  loading  &  analytics  at  any  time,  concurrently,  to   get  data  to  users  faster Scale  compute  to  support  any  workload Scale  concurrency  without  performance  impact Accelerate  the  data  pipeline
  29. 29. 29 #2 – Secure by Design with Automatic Encryption of Data! Authentication Embedded   multi-­factor  authentication Federated  authentication   available Access  control Role-­based  access   control  model Granular  privileges  on  all   objects  &  actions Data  encryption All  data  encrypted,  always,   end-­to-­end Encryption  keys  managed   automatically External  validation Certified  against  enterprise-­ class  requirements   HIPPA  Certified!
  30. 30. 30 #1  -­ Automatic  Query  Optimization •Fully  managed  with  no  knobs  or  tuning  required •No  indexes,  distribution  keys,  partitioning,  vacuuming,… •Zero  infrastructure  costs •Zero  admin  costs
  31. 31. 31 Other  Cloud  Data  Warehouse  Offerings
  32. 32. 32 Amazon  Redshift •Amazon's  data  warehousing  offering  in  AWS   • First  announced  in  fall  2012  and  GA  in  early  2013 • Derived  ParAccel  (Postgres)  moved  to  the  cloud •Pluses • Maturity: based  on  Paraccel,  on  the  market  for  almost  10  years.   • Ecosystem: Deeper  integration  with  other  AWS  products   • Amazon backing
  33. 33. 33 Amazon  Redshift •Challenges  (vs  Snowflake) • Semi-­structured  data: Redshift  cannot  natively  handle  flexible-­ schema  data  (e.g.  JSON,  Avro,  XML)  at  scale. • Concurrency: Redshift  architecture  means  that  there  is  a  hard   concurrency  limit  that  cannot  be  addressed  short  of  creating  a  second,   independent  Redshift  cluster. • Scaling: Scaling  a  cluster  means  read-­only  mode  for  hours  while  data   is  redistributed.  Every  new  cluster  has  a  complete  copy  of  data,   multiplying  costs  for  dev,  test,  staging,  and  production  as  well  as  for   datamarts  created  to  address  concurrency  scaling  limitations. • Management  overhead: Customers  report  spending  hours,  often   every  week,  doing  maintenance  such  as  reloading  data,  vacuuming,   updating  metadata.
  34. 34. 34 Customer  Analysis  – Snowflake  vs   Redshift •Ability  to  provision  compute  and  storage  separately  – •Storage  might  grow  exponentially  but  compute  needs  may  not   •No  need  to  add  nodes  to  cluster  to  accommodate  storage •Compute  capacity  can  be  provisioned  during  business  hours   and  shut  down  when  not  required  (saving  $$$) •Predictable/  exact  processing  power  for  user  queries  with   dedicated  separate  warehouses •No  concurrency  issues  between  warehouses •No  constraints  on  completing  the  ETL  run  before  business  hours   •Analytical  workload  and  ETL  can  run  in  parallel
  35. 35. 35 Customer  Analysis •0  maintenance/  100%  managed   •No  tuning  (distkey,  sortkey,  vacuum,  analyze) •100%  uptime,  no  backups •Can  restore  at  transaction  level  through  time  travel  feature •3X-­ 5X   better  compression  compared  to  Redshift •Auto  compressed •Data  at  rest  is  encrypted  by  default •With  Redshift,  performance  is  degraded  by  2x-­3x
  36. 36. 36 Customer  Analysis •Supports  cross  database  joins •With  cloning  feature,  we  can  spin  up  Dev/  test   by   cloning  entire  prod  database  in  seconds  → run  tests→ and  drop  the  clone •There  is  no  charge  for  a  clone.  Only  incremental  updates  on   storage  are  charged •Instant  Resize  (scale  up)   •NO  20+  hrs  read  only  mode  like  Redshift!   •Resize  also  allows  provisioning  higher  compute  capacity  for  faster   processing  when  required
  37. 37. 37 Microsoft  Azure  DW •Based  on  Analytics  Platform  System   •Emerging  out  of  MSFT’s  on-­premises  MPP  data  warehouse  DataAllegro   acquisition  in  2008 •Pluses • Maturity:   very  mature  (on-­prem)  database  for  over  20  years   • Ecosystem:   Deep  integration  with  Azure  and  SQL  server  ecosystem • Separation  of  compute  and  storage:  can  scale  compute  without   the  need  of  unloading/loading/moving   the  underlying  at  the  database   level.   • Integration  w/  Big  Data  &  Hadoop: allows  querying  'semi-­ structured/unstructured'  data,  such  as  Hadoop  file  formats  ORC,  RC,   and  Parquet.  This  works  via  the  external  table  concept
  38. 38. 38 Microsoft  Azure  DW •Challenges  (vs  Snowflake) • Concurrency: Azure  architecture  means  there  are  hard  concurrency   limit  (currently  32  users)  per  DWU  (Data  Warehouse  Units)  that  cannot   be  addressed  short  of  creating  a  second  warehouse  and  copy  of  the   data. • Scaling: Cannot  scale  clusters  easily  and  automatically. • Management  overhead: Azure  is  difficult  manage,  specially  with   considerations  around data  distribution,  statistics,  indices,  encryption,   metadata  operations,  replication  (for  disaster  recovery)  and  more. • Security:  lack  of  end  to  end  encryption.   • Support  for  modern  programmability: Lacks  support  for  wide   range  of  APIs  due  to  commitments  to  its  own  ecosystem.
  39. 39. 39 Google  BigQuery •Query  processing  service  offered  in  the  Google  Cloud,  first   launched  in  2010 • Follow  up  offering  from  Dremel  service  developed  internally  for  Google   only •Pluses • Google-­scale  horsepower: Runs  jobs  across  a  boatload  of  servers.   As  a  result,  BigQuery  can  be  very  fast  on  many  individual queries. • Absolutely  zero  management: You  submit  your  job  and  wait  for  it   to  return–that's  it.  No  management  of  infrastructure,  no  management   of  database  configuration,  no  management  of  compute  horsepower,   etc.
  40. 40. 40 Google  BigQuery •Challenges  (vs  Snowflake) • BigQuery  is  not  a  data  warehouse: Does  not  implement  key  features  that   are  expected  in  a  relational  database,  which  means  existing  database   workloads  will  not  work  in  BigQuery  without  non-­trivial  change   • Performance  degradation  for  JOINs:  Limits  on  how  many  tables  can  be   joined • BigQuery  is  a  black  box: You  submit  your  job  and  it  finishes  when  it   finishes–users  have  no  ability  to  control  SLAs  nor  performance.   • Lots  of  usage  limitations: Quotas  on  how  many  concurrent  jobs  can  be   run,  how  many  queries  can  run  per  day,  how  much  data  can  be  processed  at   once,  etc. • Obscure  pricing: Prices  per  query  (based  on  amount  of  data  processed  by   a  query),  making  it  difficult  to  know  what  it  will  cost  and  making  costs  add  up   quickly • BigQuery  only  recently  introduced  full  SQL  support
  41. 41. 41 Snowflake in  Action  Today
  42. 42. 42 Simplifying  the  data  pipeline Event   Data Kafka Hadoop SQL  Database Analysts  &  BI   Tools Import   Processor Key-­value   Store Event   Data Kafka Amazon  S3 Snowflake Analysts  &  BI   Tools Scenario • Evaluating event data from various sources Pain Points • 2+ hours to make new data available for analytics • Significant management overhead • Expensive infrastructure Solution Send data from Kafka to S3 to Snowflake with schemaless ingestion and easy querying Snowflake Value • Eliminate external pre-processing • Fewer systems to maintain • Concurrency without contention & performance impact
  43. 43. 43 Simplifying  the  Data  pipeline   EDW Game  Event  Data Internal  Data Third-­party  Data Analysts  &  BI   Tools Staging noSQL   Database Existing EDW Game  Event  Data Internal  Data Third-­party  Data Analysts  &  BI   Tools SnowflakeKinesis Cleanse  Normalize  Transform 11-­24  hours 15  minutes Scenario Complex pipeline slowing down analytics Pain Points • Fragile data pipeline • Delays in getting updated data • High cost and complexity • Limited data granularity Solution Send data from Kinesis to S3 to Snowflake with schemaless ingestion and easy querying Snowflake Value • >50x faster data updates • 80% lower costs • Nearly eliminated pipeline failures • Able to retain full data granularity
  44. 44. 44 Delivering  compelling  results Simpler  data  pipeline Replace  noSQL  database  with  Snowflake  for  storing  &   transforming  JSON  event  data Snowflake: 1.5  minutes noSQL  data  base:   8  hours  to  prepare  data Snowflake: 45  minutes Data  warehouse  appliance:   20+  hours Faster  analytics Replace  on-­premises  data  warehouse  with  Snowflake   for  analytics  workload Significantly  lower  cost Improved  performance  while  adding  new  workloads-­-­at   a  fraction  of  the  cost Snowflake: added  2  new  workloads  for  $50K Data  warehouse  appliance:   $5M  +  to  expand
  45. 45. 45 The  fact  that  we  don’t  need   to  do  any  configuration  or   tuning  is  great  because  we   can  focus  on  analyzing  data   instead  of  on  managing   and  tuning  a  data   warehouse.” Craig  Lancaster,  CTO The  combination  of   Snowflake  and  Looker  gives   our  business  users  a   powerful,  self-­service  tool  to   explore  and  analyze  diverse   data,  like  JSON,  quickly  and   with  ease.” We  went  from  an  obstacle   and  cost-­center  to  a  value-­ added  partner  providing   business  intelligence  and   unified  data  warehousing   for  our  global  web  property   business  lines.” JP  Lester,  CTO Music  &  film  distribution Publishers  of,,,   and  other  premium  websites Internet  access  service  company  to over  30  million  smartphone  users • Replaced  MySQL • Substituted  Snowflake native  JSON  handling and  queries  with  SQL in  place  of  MapReduce • Integrated  with  Hadoop repository • Consolidated  global   web  data  pipelining • Replaced  36-­node  data warehouse • Replaced  100-­node   Hadoop  cluster Erika  Baske,  Head  of  BI • Eliminated  bottlenecks and  scaling  pain-­points • Consolidated  multiple cloud  data  marts • Now  handling  larger   datasets  and  higher   concurrency  with  ease
  46. 46. 46 Cloud-­scale  data  warehouse
  47. 47. 47 Steady  growth  in  data  processing •Over  20  PB  loaded  to  date! •Multiple  customers  with  >1PB   •Multiple  customers  averaging  >1M   jobs  /  week   •>1PB  /  day  processed   •Experiencing  4X  data  processing growth  over  last  six  months Jobs  /  day
  48. 48. 48 What  does  a  Cloud-­native  DW  enable? Cost  effective  storage  and  analysis  of  GBs,  TBs,  or  even  PB’s Lightning  fast  query  performance   Continuous  data  loading  without  impacting  query  performance Unlimited  user  concurrency ODBC JDBC Interfaces Java >_ Scripting Full  SQL  relational  support  of  both  structured  and   semi-­structured  data Support  for  the  tools  and  languages  you  already  use
  49. 49. 49 Making  Data  Warehousing  Great  Again!
  50. 50. 50 As  easy  as  1-­2-­3! Discover  the  performance,  concurrency,   and  simplicity  of  Snowflake 1 Visit 2 Click  “Try  for  Free” 3 Sign  up  &  register Snowflake  is  the  only  data  warehouse  built  for  the  cloud.  You  can   automatically  scale  compute  up,  out,  or  down̶—independent   of  storage.   Plus,  you  have  the  power  of  a  complete  SQL  database,  with  zero   management,  that  can  grow  with  you  to  support  all  of  your  data  and  all   of  your  users.  With  Snowflake  On  Demand™,  pay  only  for  what  you  use.   Sign  up  and  receive $400  worth  of  free usage  for  30  days!
  51. 51. Kent Graziano Snowflake Computing On  Twitter  @KentGraziano More  info  at Visit  my  blog  at Contact  Information
  52. 52. YOUR  DATA,  NO  LIMITS Thank  you