What is new in the cloud? - A Database Perspective
Transcript of What is new in the cloud? - A Database Perspective
![Page 1: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/1.jpg)
What is new in the cloud?- A Database Perspective –
Donald Kossmann
Systems Group, ETH Zurich
http://systems.ethz.ch
![Page 2: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/2.jpg)
Acknowledgments
![Page 3: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/3.jpg)
Reference
• Kossmann, Kraska: Data Management in the Cloud: Promises, State-of-the-art, and Open Questions. Datenbank Spektrum, 2010.
– http://www.pubzone.org/dblp/journals/dbsk/KossmannK10
– (Published by Springer Open Access.)
• (Summary of several talks and papers of the DB community on cloud computing.)
![Page 4: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/4.jpg)
Agenda
• Promises of Cloud Computing
• Benchmarking the State-of-the Art [SIGMOD 2010]
– Amazon, Google, Microsoft
• Towards a Silver Bullet [ICDE 2010]
– Asking the right questions
![Page 5: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/5.jpg)
The Cloud 15 Years Ago
BrowserBrowser
Internet
DB 1
Service 1
DB 2
Service 2
http http
httphttp
![Page 6: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/6.jpg)
Service 1 Service 2 Service 3
Browser Adobe AirAdobe Flex
Mobile Games ...
Internet
Internal & External Data
ClientMachines
Serversof ut ilityprovider
REST (http)
App1
Doc Doc
App1DB
Doc Doc
App1App1DB
Doc
The Cloud Today
http, udp, …
![Page 7: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/7.jpg)
Promises of Cloud Computing
• Cost
– reduce cost
• utilization, commoditization, optimization
• prevention of failures, having failures
– „pay as you go“ (cap-ex vs. op-ex)
– easy to test!
• Time to market
– avoid unnecessary steps
• HW provisioning, puchasing, testing
– difficult to test!
• can only test scalability / elasticity
![Page 8: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/8.jpg)
Cow vs. Supermarket
Cow Supermarket: Bottle of Milk
Big upfront investment Small continous expense
Produces more than you need Buy what you need
Uses up resources Less resources needed
Maintenance needed No maintenance
Unpleasant waste product Waste is not my problem
[Abadi 2009], [Kraska 2009]
![Page 9: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/9.jpg)
Questions to ask
• Questions you ask your supermarket
– How expensive is one litre of milk?
– Is the price always the same?
– What if I have a big party? Can you deliver?
• Questions you do not ask
– How fast is the supermarket?
– Good enough is good enough!
![Page 10: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/10.jpg)
What to optimize?
Feature Traditional Cloud
Cost [$] fixed optimize
Performance [tps, secs] optimize fixed
Scale-out [#cores] optimize fixed
Predictability [s($)] - fixed
Consistency [%] fixed ???
Flexibility [#variants] - optimize
[Florescu & Kossmann, SIGMOD Record 2009]
Put $ on the y-axis of your graphs!!!
![Page 11: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/11.jpg)
Agenda
• Promises of Cloud Computing
• Benchmarking the State-of-the Art [SIGMOD 2010]
– Amazon, Google, Microsoft
• Towards a Silver Bullet [ICDE 2010]
• Asking the right questions
![Page 12: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/12.jpg)
What is out there?
• Amazon
– IaaS: S3 and EC2
– PaaS: SQS, SimpleDB, RDS
• Google App Engine
– PaaS: Integrated DB, AppServer, and Cache
• Microsoft Azure
– PaaS: Windows Azure, SQL Azure, .Net Azure
• Gazillions of start-ups (e.g., 28msec )
• Research prototypes (e.g., Crescando )
![Page 13: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/13.jpg)
What is out there?
• Amazon
– IaaS: S3 and EC2
– PaaS: SQS, SimpleDB, RDS
• Google App Engine
– PaaS: Integrated DB, AppServer, and Cache
• Microsoft Azure
– PaaS: Windows Azure, SQL Azure, .Net Azure
• Gazillions of start-ups (e.g., 28msec )
• Research prototypes (e.g., Crescando )
![Page 14: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/14.jpg)
Scope of this study
• Workloads: Focus on OLTP
– OLAP under heavy debate by others
– streaming not addressed yet (~ OLTP)
– testing, archiving, etc. is boring
• Types of clouds: Focus on public clouds
– believe that applicable to any kind of cloud
• only difference: private clouds have planned downtime
– cloud on the chip
– swarms: ad-hoc private clouds
• IaaS vs. PaaS vs. SaaS: Focus on PaaS
![Page 15: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/15.jpg)
Tested Services
MS Azure Google App Eng AWS RDS AWS S3
Business Model PaaS PaaS PaaS IaaS
Architecture Replication Part. + Repl.
(+Dist. Control)
Classic Distr.
Control
Consistency SI ≈ SI Rep. Read EC
Cloud Provider Microsoft Google Amazon Flexible
Web/App
Server
.Net Azure AppEngine Tomcat Tomcat
Database SQL Azure DataStore MySQL --
Storage / FS Simple DataStore GFS -- S3
App-Language C# Java/AppEngine Java Java
DB-Language SQL GQL SQL Low-Lev.
API
HW Config. Part. automatic Automatic Manual Manual
![Page 16: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/16.jpg)
Tested Services
AWS SimpleDB AWS MySQL AWS MySQL/R
Business Model PaaS IaaS IaaS
Architecture Replication Classic Replication
Consistency EC Rep. Read Rep. Read
Cloud Provider Amazon Flexible Flexible
Web/App Server Tomcat Tomcat Tomcat
Database SimpleDB MySQL MySQL
Storage / File
System
-- EBS EBS
App-Language Java Java Java
DB-Language SimpleDB Queries SQL SQL
HW Config Partly Automatic Manual Manual
![Page 17: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/17.jpg)
Benchmark Setup
VM
MySQL
VM
RDSRDSMySQL R (VM)
VM
SimpleDB
VM
S3
.Net
MS SQL Azure
VM
MySQL
MySQL
R
Simple
DBS3
DataCenter B
DataCenter AVM VM VM VM VM
Rem
ote
Bro
wser
Em
ula
tor (R
BE
)S
yste
m U
nde
r Te
st (S
UT
)
App Engine
Data Store
VM
RDS
RDSEBS
Manually mgt, manually scaled Provider mgt, manually scaled Provider mgt, provider scaled
![Page 18: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/18.jpg)
Benchmarking Cloud Services
• Goal: Test if cloud promises are fulfilled
• Benchmarking Approach [DBTest09, SIGMOD10]
– TPC-W Benchmark, Ordering Mix
– Adapted for testing elasticity and cost ($)
– Vary load (EB): clicks per second
• Benchmark metrics
– Cost: $ / WI
– Cost Predictability: s($/WI)
– Throughput: WIPS
![Page 19: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/19.jpg)
EBs Fully
Utilized1 10 100 500 1000
MySQL 0.635 0.072 0.020 0.006 0.006 0.005
MySQL/R 2.334 0.238 0.034 0.008 0.006 0.005
RDS 1.211 0.126 0.032 0.008 0.006 0.005
SimpleDB 0.384 0.073 0.042 0.039 0.037 0.037
S3 1.304 0.206 0.042 0.019 0.011 0.009
Google AE 0.002 0.028 0.033 0.042 0.176 0.042
Google AE/C 0.002 0.018 0.026 0.028 0.134 0.028
Azure 0.775 0.084 0.023 0.006 0.006 0.005
Cost [m$/WI]
![Page 20: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/20.jpg)
Cost predictability [m$/WI]
• Ideal: Cost / WI is constant
– low s better (ideal: s=0)
– cost independent of load
• Google clear winner here!
mean± s
MySQL 0.015± 0.077
MySQL/R 0.043± 0.284
RDS 0.030± 0.154
SimpleDB 0.063± 0.089
S3 0.018± 0.098
Google AE 0.029± 0.016
Google AE/C 0.021± 0.011
MS Azure 0.010± 0.058
![Page 21: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/21.jpg)
Relative Cost FactorsMySQL Azure
SimpleDB S3
23
![Page 22: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/22.jpg)
Throughput [WIPS]
0
100
200
300
400
500
600
700
800
900
1000
1100
1200
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
WIP
S
Emulated Browsers
SimpleDB
GoogleAE
S3
Azure
RDS
MySQL
Ideal
![Page 23: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/23.jpg)
0
50
100
150
200
0 500 1000 1500 2000 2500
WIP
S
Emulated Browsers
0
200
400
600
800
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
WIP
S
Emulated Browsers
0
200
400
600
800
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
WIP
S
Emulated Browsers
Overload BehaviorRDS SimpleDB
App Engine
![Page 24: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/24.jpg)
Agenda
• Promises of Cloud Computing
• Benchmarking the State-of-the Art [SIGMOD 2010]
– Amazon, Google, Microsoft
• Towards a Silver Bullet [ICDE 2010]
– Asking the right questions
![Page 25: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/25.jpg)
Principles of Distributed Databases
• Partitioning [Ceri,Pelagatti 1984]
– Large number of possible partitioning schemes
– Repartitioning is very expensive
• Replication, Caching [Bernstein et al. 1987]
– Replicating data increases fault-tolerance and read performance
– Replication needs a mechanism to keep replicas consistent
• Distributed Control [Tanenbaum 2002]
– Client-side consistency protocols
• Combination of approaches difficult! [Unterbrunner+ 2010]
![Page 26: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/26.jpg)
Reference Architecture
Client
Store
HTTP
Web Server
App Server
DB Server
FCGI, ...
SQL
get/put block
records
XML, JSON, HTML
XML, JSON, HTML
![Page 27: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/27.jpg)
Open Questions
Client
Store
Web Server
App Server
DB Server
• How to map stack to IaaS?
• How to implement store layer?
• What consistency model?
• What programming model?
• Whether and how to cache?
![Page 28: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/28.jpg)
Open Questions
Client
Store
Web Server
App Server
DB Server
• How to map stack to IaaS?
• How to implement store layer?
– [VLDB 2009a]
• What consistency model?
– [VLDB 2009b]
• What programming model?
– [VLDB 2009c]
• Whether and how to cache?
– [SIGMOD 2010]
![Page 29: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/29.jpg)
Variant I: Partition Workload by „Tenant“
Client
Store
HTTP
Web Server
App Server
DB Server
FCGI, ...
SQL
get/put block
records
XML, JSON, HTML
XML, JSON, HTML
Client ClientClient
Workload Splitter
Store-A
Server-A
Server-A
Server-A
Store-B
Server-B
Server-B
Server-B
XML, JSON, HTML
block
![Page 30: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/30.jpg)
Partition Workload by „Tenant“• Principle
– partition data by „tenant“
– route request to DB of that tenant
• Advantages
– reuse existing database stack (RDBMS)
– flexibility to use DAS or SAN/NAS
• Disadvantages
– multi-tenant problem [Salesforce]
• optimization, migration, load balancing, fix cost
– silos: need DB federator for inter-tenant requests
– expensive HW and SW for high availabilty
![Page 31: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/31.jpg)
Metaphor: Shopping Mall
• If a shop is successful, you need to move it!• (popularity vs. growth of product assortment)
![Page 32: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/32.jpg)
Variant II: Partition Workload by „Request“
Client
Store
HTTP
Web Server
App Server
DB Server
FCGI, ...
SQL
get/put block
records
XML, JSON, HTML
XML, JSON, HTML
Client ClientClient
Workload Splitter
Store (e.g., S3)
Server-A Server-B
XML, JSON, HTML
???
Store (e.g., S3)Store (e.g., S3)
![Page 33: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/33.jpg)
Metaphor: Internet Department Store
• If a product is successful, you stock up its supply
– Transparent and fine-grained reprovisioning
– Cost of reprovisioning much lower!!!
Server-AServer-A
Server-AServer
Store (e.g., S3)Store (e.g., S3)
Store (e.g., S3)
![Page 34: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/34.jpg)
Partition Workload by „Request“
• Principle
– fine-grained data partitioning by page or object
– any server can handle any request
– implement DBMS as a library (not server)
• Advantages
– avoids disadvantages of Variant I
• Disadvantages
– new synchronization problem
– whole new breed of systems
– caching not effective
![Page 35: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/35.jpg)
Experiments Revisited: Cost / WI (m$)
Low Load Peak Load
Variant 1 (RDS) 1.212 0.005
Variant 2 (S3) - 0.007
Variant 2 (Google) 0.002 0.028
Variant 1 (Azure) 0.775 0.005
Only V2 is cheap and predictable!�
![Page 36: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/36.jpg)
Throughput [WIPS]
0
100
200
300
400
500
600
700
800
900
1000
1100
1200
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
WIP
S
Emulated Browsers
SimpleDB
GoogleAE
S3
Azure
RDS
MySQL
Ideal
V1, V2
V1
V1, V2
High performance can be achieved with both!Warning: Still a hypothesis. S3 is only EC.
![Page 37: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/37.jpg)
Open Questions
• How to map traditional DB stack to IaaS?
• How to implement the storage layer?
• What is the right consistency model?
• What is the right programming model?
• Whether and how to make use of caching?
![Page 38: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/38.jpg)
Storage Layer Questions
Client
Store
Web Server
App Server
DB Server
get/put block
• What is the interface?
– get/put or more?
– return records or blocks?
• How to implement this interface?
– data structures
– storage media and network
– clustering of data
– granularity of partitioning / sharding
![Page 39: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/39.jpg)
API of Store• Assumption: Hardware trends will change the game
– most data in MM or SSD / PCM
• Proposition 1: return tuples rather than blocks
– blocks are an arte-fact of old HW (i.e., disks)
• Proposition 2: push down predicates
– because it can be done (power depends on impl.)
– compensates for cost of remote storage
– (old idea; e.g., iDisks)
• Open: consistency constract of store?
– store might have nice properties (e.g., mon. Writes)
– store might have nice primitives (e.g., counter, set/test)
![Page 40: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/40.jpg)
Store Implementation Variants• DAS
– local disks with physically exclusive access
– put/get interface; no synchronization
– only works for V1
• Key-value stores (e.g., Dynamo)– DHTs with concurrent access
– put/get interface; no synchronization
– works for V1 and V2; makes more sense for V2
• A DBMS (e.g., MySQL, SQL Server)– have been misused before
• ClockScan [Unterbrunner et al. 2009]
– massively shared scans in a distributed system
![Page 41: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/41.jpg)
Open Questions
• How to map traditional DB stack to IaaS?
• How to implement the storage layer?
• What is the right consistency model?
• What is the right programming model?
• Whether and how to make use of caching?
![Page 42: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/42.jpg)
CAP Theorem
• Three properties of distributed systems
– Consistency (ACID transactions w. serializability)
– Availability (nobody is ever blocked)
– resilience to network Partitioning
• Result
– it is trivial to achieve 2 out of 3
– it is impossible to have all three
• Highly controversial discussion [Brantner 08, Abadi 10]
– Levels of availability (always vs. never available)
– What is the difference between CP and CA
![Page 43: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/43.jpg)
What have people done?• Client-side Consistency Models [Tannenbaum]
– read & write monotonicity, session consistency, ...
• Time-line consistency [PNUTS08]
– all writes are consistent (except across DC)
– read consistency: either don‘t care or latest version
• New DB transaction models
– Escrow, Reservation Pattern [O‘Neil 86], [Gawlick 09]
– SAGAs and compensation; e.g., in BPEL [G.-Molina,Salem]
– SAP, Amadeus et al. [Buck-Emden], [Kemper et al. 98]
• Define logical unit of consistency [Azure]
– strong consistency within LuC
– app-defined consistency across LuCs
• Educate Application Developers [Helland 2009]
![Page 44: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/44.jpg)
What have people done?• Client-side Consistency Models [Tannenbaum]
– read & write monotonicity, session consistency, ...
• Time-line consistency [PNUTS08]
– all writes are consistent (except across DC)
– read consistency: either don‘t care or latest version
• New DB transaction models
– Escrow, Reservation Pattern [O‘Neil 86], [Gawlick 09]
– SAGAs and compensation; e.g., in BPEL [G.-Molina,Salem]
– SAP, Amadeus et al. [Buck-Emden], [Kemper et al. 98]
• Define logical unit of consistency [Azure]
– strong consistency within LuC
– app-defined consistency across LuCs
• Educate Application Developers [Helland 2009]
Sacrifice ConsistencyPush problem to app programmer!
![Page 45: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/45.jpg)
What have we done?
• Levels of Consistency, Cost Tradeoffs [Brantner08]
– can achieve read/write monotonicy + „A“ + „P“
– the more consistency, the higher the cost
• Economic models for consistency [Amadeus], [Kraska09]
– Classify the data as A, B, C
• A data always strong consistent
• C data always eventually consistent
• B data handled like A or C data adaptively
– Cost model that estimates business impact of inconst.
![Page 46: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/46.jpg)
What have we done?
• Levels of Consistency, Cost Tradeoffs [Brantner08]
– can achieve read/write monotonicy + „A“ + „P“
– the more consistency, the higher the cost
• Economic models for consistency [Amadeus], [Kraska09]
– Classify the data as A, B, C
• A data always strong consistent
• C data always eventually consistent
• B data handled like A or C data adaptively
– Cost model that estimates business impact of inconst.
Follow tradition topush problem to app programmer!
But for a different reason!!!
![Page 47: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/47.jpg)
Cost per 1000 TAs [$] (TPC-W)
Total Adjustable Fixed
Nothing 0.15 0 0.15
Durability 1.8 1.1 0.7
Monotonicity 2.1 1.4 0.7
Monotonicity+Atomicity 2.9 2.6 0.3
[Brantner+08]
![Page 48: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/48.jpg)
Latency per TAs [secs] (TPC-W)
Avg. Max.
Nothing 11.3 12.1
Durability 4.0 5.9
Monotonicity 4.0 6.8
Mono+Atomicity 2.8 4.6
[Brantner+08]
![Page 49: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/49.jpg)
Open Questions
• How to map traditional DB stack to IaaS?
• How to implement the storage layer?
• What is the right consistency model?
• What is the right programming model?
• Whether and how to make use of caching?
![Page 50: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/50.jpg)
Programming Model
• Properties of a programming lang. for the cloud
– support DB-style + OO-style + CEP-style
– avoid keeping state at app servers for V2
• Many languages will work in the cloud
– SQL, XQuery, Ruby, .Net /LINQ, ...;
– J2EE will not work
• Open (research) questions
– do OLAP on the OLTP data: My guess is yes!
– rewrite your apps: My guess is yes!
![Page 51: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/51.jpg)
Caching
• Many Variants Possible
– this is just one
– V1 caching mandatory
– V2 caching prohibitive
• TPC-W Experiments
– marginal improvements for Google AppEngine
• No low hanging fruit
Web Server +
App Server
Clients
DB Server
Storage
Traditional
Clients
Web Server +
App Server +
DB Server
Storage
S3
Web Server+
App Server
Clients
MS Azure
DB Server +
Storage
SQL
Put/Get
SQL
App Server +
Web Server
Clients
MemCache
DB Server
Storage
Google App Cache
SQL Put/Get
Put/Get
Put/Get
Clients
Web Server +
App Server +
Database
Storage
S3 Alternative
Put/Get
TRX Services
Web Server+
App Server
Clients
DB Server +
Storage
SQL
HTTP
HTTP
HTTP
HTTP
HTTP
HTTP
![Page 52: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/52.jpg)
Agenda
• Promises of Cloud Computing
• Benchmarking the State-of-the Art [SIGMOD 2010]
– Amazon, Google, Microsoft
• Towards a Silver Bullet [ICDE 2010]
– Asking the right questions
![Page 53: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/53.jpg)
Case Study: Bets Made by 28msec
• How to map traditional DB stack to IaaS?
– implemented both architectures (V1 + V2)
– V1 only in a single server variant for low end
• How to implement the storage layer?
– EBS for V1; KVS for V2
• What is the right consistency model?
– ACID for V1; configurable for V2
• What is the right data + programming model?
– XML & XQuery
• Whether and how to make use of caching?
– No! (Only for code / precompiled query plans)
![Page 54: What is new in the cloud? - A Database Perspective](https://reader035.fdocuments.in/reader035/viewer/2022071600/613d2227736caf36b759b5d9/html5/thumbnails/54.jpg)
Conclusion & Future Work
• Hopefully started a new benchmark war
– Tested elasticity, cost and cost predictability
– Vendors interested in cost experiments
– Academia interested in scalability experiments
– More bits available: http://www.pubzone.org
• Find silver bullet for data management in cloud
– Develop a reference architecture
– Combine partitioning + replication + distributed control
– ETH Systems Group: http://www.systems.ethz.ch