mining for insight - o'reilly mediaassets.en.oreilly.com/1/event/21/mining for insight within...
TRANSCRIPT
![Page 1: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/1.jpg)
Mining for insightOsma Ahvenlampi, CTO, Sulake
Implementing business intelligence for Habbo
![Page 2: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/2.jpg)
Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
![Page 3: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/3.jpg)
3Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Virtual world
![Page 4: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/4.jpg)
4Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Social Play
![Page 5: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/5.jpg)
5Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Habbo Countries
![Page 6: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/6.jpg)
6Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Leading virtual world
2009
6
20052000 2002 2003 20042001 20082006
129Million
11.7 Million/month!
2007
» 129 million registered Habbo-characters› Source: Sulake Statistics, March 2009
» 11.7 million unique browsers per month› Source: Google Analytics, March 2009
» 2 million visits / day
» 40 million hours of play /
month
![Page 7: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/7.jpg)
7Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Overview
• Analytics approach and objectives• Types of data processing• Description of a solution for scaling event storage
and analysis• Observations about Infobright technology
![Page 8: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/8.jpg)
Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Background
![Page 9: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/9.jpg)
9Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Scaling a virtual world
• Java code is “easy” to scale– Clustered, load-balanced process model on J2SE +
open source stack
• MySQL not so much
• Local communities provide natural shards
![Page 10: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/10.jpg)
10Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Data management in Habbo
• Several dozen DB servers• Close to a hundred MySQL processes
– MySQL on many-core hardware!
• Terabytes of managed data– Fragmented all over the place
• 3 million new user accounts monthly• 2 million visits daily, average ~40 minutes• Hundreds of interactions every visit• Hundreds of millions of user-created “rooms”
![Page 11: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/11.jpg)
Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Analytics
![Page 12: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/12.jpg)
12Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Business objective
• Development process is iterative– Benefits from constant learning
• Requires data– Users, visits, meetings, purchases, trading, friends,
events, activities, achievements, places, items and so on..
• Up-to-date management information• Virtual worlds == virtual economies
– Economies require oversight
![Page 13: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/13.jpg)
13Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Examples of analysis
• Spending patternshttp://bit.ly/B8sg
• Behavioral segments
Casual visi-tors
Regular users Customers High spenders
0%
10%
20%
30%
40%
50%
60%
Accounts Time spent
![Page 14: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/14.jpg)
Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Managing dataHow to operate, collect, and analyse data
at Habbo's scale
![Page 15: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/15.jpg)
15Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Three types of data processing
Real-time shared state• In-memory data structures
• Game (business) logic
• Try to keep data footprint small
Log files & analytics• High-volume events
• Post-processing
OLTP• Transactional integrity
• Persistent customer state
![Page 16: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/16.jpg)
Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Event logs and analytics w/ MySQLRecap of methods we've used over time
![Page 17: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/17.jpg)
17Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Large analytics tables and MyISAM
• Fast writes, as long as you don't maintain a lot of indexes
• Fast reads, but only if you do maintain a lot of indexes
• Terrible crash behavior– Have you ever tried to myisamchk a 1 TB table?
• Good for interim or throw-away buffers only
![Page 18: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/18.jpg)
18Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Large analytics tables and InnoDB
• Ok for OLTP• Crash-safe• Pretty slow for batch loads, even after lots of tuning
– Google and Percona patches help!
• Not good for complex tables with lots of indexes• Horrible if you ever need to change your table
schema– analysis databases change constantly!
![Page 19: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/19.jpg)
19Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Take 2
• Must be able to load millions of rows every day• Retaining billions of rows• Schema evolves with new features and improving
analysis• Can't afford days of downtime for changes or
maintenance
![Page 20: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/20.jpg)
20Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Columnar databases
• Turn storage by 90 degrees• Enables very wide tables and rapid access to narrow
sets of columns• Compresses well• Perfect for data warehousing• Not a new field, but enjoying a comeback
– Expensive MPP solutions; Vertica, ParAccel..
• For MySQL, there's at least Infobright
![Page 21: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/21.jpg)
21Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Applications for columnar storage
• Data Warehousing / Business Intelligence– Next-day results (typically)– Big storage, complex data models, lots of repetition– Analytic query performance
• Event log management– Not realtime, but as close to it as possible– Very high volumes, simple content– Long-term storage issue
![Page 22: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/22.jpg)
Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
An approach to loggingHigh-performance data collection
![Page 23: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/23.jpg)
23Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Structures of logfiles
• “Clickstream” event dataflows– Logins, logouts, messages, actions– For unstructured or semi-structured-data, Hadoop
• Simple structure, don't even try to be relational– Immediate output– Tradionally on small scale with text files or MyISAM
• Direct analysis is tricky– Long-lasting activities split to “begin” and “end”– What if there's an interruption?– Related events scattered around
![Page 24: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/24.jpg)
24Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Overall architecture http://bit.ly/ice-logs
1. Java clients buffer locally and sends batch data over RMI to log server
2. Log servers buffers to local interim flat file
3. Log loader takes new files and loads them to an Infobright ICE database
4. Files are then removed
Single-thread performance using low-end hardware: 100,000 processed log entries per second
![Page 25: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/25.jpg)
25Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Keep it simple
• Don't complicate event processing by making logging structures more complex
• Reasons to keep the format simple:– It's simple to implement– Eventually, scale will require it anyway
• Process the data to a richer structure asynchronously– Eg, using Hadoop
![Page 26: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/26.jpg)
Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
From logging to data warehousingCollection is nice, but using the data is nicer
![Page 27: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/27.jpg)
27Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Analysis tools
• Raw event streams are difficult to use as-is• Multiple sources for information(!)• Postprocess and integrate
– Combine or link related events– Calculate value(s)
• Store in a schema which facilitates dimensional reporting (eg, star or snowflake)
![Page 28: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/28.jpg)
28Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Data warehousing process
Datasources
Stage &extract
Transformations
StarSchema
DWH
1. Extract data from all sources, whether external, OLTP databases, or event logs
2. Identify common dimensions, related data, transform structure and reorganize schema
3. Load to the final data warehouse
![Page 29: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/29.jpg)
29Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Tools for data integration
• Could be scripted, but maintenance is a killer– Documentation– Data processing nasty to deal with in script form
• Choose a tool from the start– Pentaho/Kettle– Talend– BO DataIntegrator– Informatica– Etc
![Page 30: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/30.jpg)
30Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Reporting and analysis
• Handwritten SQL– Expressive but cumbersome
• OLAP cubes– Rapid but memory limited, require constant reloads
• Query builders– With a good UI and DB schema, it's what I'd choose
• Specialized tools where available– For web traffic analysis, building anything seems mad
when tools like Google Analytics available for free
![Page 31: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/31.jpg)
Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Columnar storage in MySQLObservations on our Infobright solution
![Page 32: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/32.jpg)
32Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Infobright's Brighthouse
• Columnar engine for MySQL– Its own server (5.1 based), not a pluggable engine
• No indexing of data required– Data is packed per-column per 64k row values– Engine maintains summary data per each 64k values– Queries target each pack where summary matches– Joins are supported by additional pack-to-pack data– Ideal for numeric data
• Fast data loader• Most MySQL-compatible tools “just work”
![Page 33: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/33.jpg)
33Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Immediate benefits
• 4 times faster loads (without tuning rest of out toolchain)
• 1/8th of the disk space needed, so we could reallocate terabytes of storage to other uses
• No time spent worrying about ad-hoc business report needing a nonexistent index
• Typical query performance increase significant
![Page 34: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/34.jpg)
34Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Queries which benefit a lot
• Summarizing of big tables w/ GROUP BY– Historically a horrible area for MySQL (requires disk-
based temporary tables and sorts)– Infobright often executes from the knowledge grid
information only, without even accessing the actual data!
• These can be tens or hundreds of times faster
select t2.c3,sum(t1.c1) from t1 join t2 using (c2) where c2 between 1 and 1000000 group by c3
![Page 35: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/35.jpg)
35Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Unsuitable applications
• Single-row selects aren't Infobright's natural domain– Row-based engine would fetch by index– Infobright needs to unpack 64k rows from each
columnar datapack
• Typically execute in about 1 second (instead of milliseconds)
select * from t1 where pk=?
• Text searching
![Page 36: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/36.jpg)
36Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Limitations
• You might be used to functionality which doesn't work with the engine– No DML in community edition (insert,update,delete)– Practical restrictions to DML also in enterprise edition
• No constraints, primary keys or auto increment• Table changes still require a full reload• Diagnostics are under-developed
– No useful query plan “explain”– Query states give little insight (always “init”)– No engine status internals exposed
![Page 37: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/37.jpg)
37Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Differences in query use
• Mostly, the same query strategies as with normal MySQL, however some notable differences
• Joins perform well when pack-to-pack metadata is available– Eg, FROM t1 JOIN t2 ON t1.column1=t2.column2– No arithmetic operations in join conditions– No BETWEEN or inequality operators (in joins!)– LEFT JOIN is sometimes fast, many times slow
• On the other hand, WHERE .. IN (subselect) usually is very quick
• Every column selected adds cost (more I/O)
![Page 38: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/38.jpg)
38Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Sizing and scaling considerations
• Table-level locking impacts loading strategy– SELECT vs INSERT/LOAD starvation issues similar to
MyISAM (w/ “low priority updates”)
• Parallel loader, but no parallelizing select• Can't replicate or partition over multiple servers• Practical scaling assumptions:
– one host server– 500 times main memory in data storage (less actual
disk utilized thanks to compression)– # of cores equal to # of users– Eg; 30 TB on a single HP DL 585 compressed to 6 TB
![Page 39: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/39.jpg)
39Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Different tech for different purposes
Real-time shared state• In-memory data structures
• Java clusters, Memcached, etc
• Future: MySQL Cluster NDB?
Log files & analytics• High-volume events
• Infobright ICE / IEE
OLTP• Transactional integrity
• InnoDB
![Page 40: Mining for insight - O'Reilly Mediaassets.en.oreilly.com/1/event/21/Mining for Insight Within Two... · Mining for insight Osma Ahvenlampi, CTO, Sulake Implementing business intelligence](https://reader031.vdocuments.mx/reader031/viewer/2022021823/5b4fbad07f8b9a166e8d00c3/html5/thumbnails/40.jpg)
Osma Ahvenlampi - MySQL Conf 2009 - Mining for insight
Thank you!
www.sulake.comwww.habbo.com
[email protected]/osma