Answering the Database Scale Out Problem: SSDs in the Data Center April 14, 2010
description
Transcript of Answering the Database Scale Out Problem: SSDs in the Data Center April 14, 2010
![Page 1: Answering the Database Scale Out Problem: SSDs in the Data Center April 14, 2010](https://reader038.fdocuments.us/reader038/viewer/2022110104/56815dd8550346895dcc0333/html5/thumbnails/1.jpg)
Answering the Database Scale Out Problem:
SSDs in the Data CenterApril 14, 2010
Dan MarriottDirector - Production Operations
![Page 2: Answering the Database Scale Out Problem: SSDs in the Data Center April 14, 2010](https://reader038.fdocuments.us/reader038/viewer/2022110104/56815dd8550346895dcc0333/html5/thumbnails/2.jpg)
Answers.com
• The world’s leading Q&A site
• Rank in top Web properties #18 in the U.S. (02/2010)(1)
#31 worldwide (02/2010) (1)
• Unique monthly visitors 50 million in the U.S. (02/2010)(1)
72 million worldwide (02/2010) (1)
2
(1) Source: comScore – Hybrid Measurement Methodology (U.S. only) beginning August 2009
![Page 3: Answering the Database Scale Out Problem: SSDs in the Data Center April 14, 2010](https://reader038.fdocuments.us/reader038/viewer/2022110104/56815dd8550346895dcc0333/html5/thumbnails/3.jpg)
ReferenceAnswers
3
![Page 4: Answering the Database Scale Out Problem: SSDs in the Data Center April 14, 2010](https://reader038.fdocuments.us/reader038/viewer/2022110104/56815dd8550346895dcc0333/html5/thumbnails/4.jpg)
WikiAnswers: Q&A the Wiki Way
4
![Page 5: Answering the Database Scale Out Problem: SSDs in the Data Center April 14, 2010](https://reader038.fdocuments.us/reader038/viewer/2022110104/56815dd8550346895dcc0333/html5/thumbnails/5.jpg)
Database layer
• MySQL 5.0.87-b20-percona• MySQL 5.0.51a
5
![Page 6: Answering the Database Scale Out Problem: SSDs in the Data Center April 14, 2010](https://reader038.fdocuments.us/reader038/viewer/2022110104/56815dd8550346895dcc0333/html5/thumbnails/6.jpg)
Challenges
• Keep site fast while site traffic and stored data are ever-
increasing
• Replication lag = 0 or users get stale data
• Forever being forced to further optimize queries constantly vying for dev resources to do this
• Controlling hardware growth (Cap & OpEx $$$) regularly adding servers to handle growth
6
![Page 7: Answering the Database Scale Out Problem: SSDs in the Data Center April 14, 2010](https://reader038.fdocuments.us/reader038/viewer/2022110104/56815dd8550346895dcc0333/html5/thumbnails/7.jpg)
Handling high growth – database tier
• Separate reads and writes• Add more read DB slaves• Use Memcached where possible• Optimize Queries• Partition large databases
_________
Started hitting a wall: Replication Lag even when servers handling modest # queries/sec
7
![Page 8: Answering the Database Scale Out Problem: SSDs in the Data Center April 14, 2010](https://reader038.fdocuments.us/reader038/viewer/2022110104/56815dd8550346895dcc0333/html5/thumbnails/8.jpg)
Typical DB read cluster
8
![Page 9: Answering the Database Scale Out Problem: SSDs in the Data Center April 14, 2010](https://reader038.fdocuments.us/reader038/viewer/2022110104/56815dd8550346895dcc0333/html5/thumbnails/9.jpg)
Fusion-io for HP Blade Servers
• March ’09: HP announce IO Accelerator card for blades (manuf. by Fusion-io)
• Sizes: 80 & 160GB SLC 320GB MLC
• April ’09: received two cards – began testing
9
![Page 10: Answering the Database Scale Out Problem: SSDs in the Data Center April 14, 2010](https://reader038.fdocuments.us/reader038/viewer/2022110104/56815dd8550346895dcc0333/html5/thumbnails/10.jpg)
Easy to install
• One man job. Takes 60 secs.
10
![Page 11: Answering the Database Scale Out Problem: SSDs in the Data Center April 14, 2010](https://reader038.fdocuments.us/reader038/viewer/2022110104/56815dd8550346895dcc0333/html5/thumbnails/11.jpg)
Performance Tests
TestBlade
Server –SAS HDDs
Blade Server –
Fusion-io card
Improvement
11
Additionally, CPU load dropped from 30% to 18% (even with Fusion-io driver overhead)
Replication catch-up time (after restore)
> 6 hours 12½ mins 3,000%
Max Queries/sec(Seconds_Behind_Master: 0)
350 Q/sec 3,500 Q/sec 900%Application response time 100 ms 70 ms 30%
Full DB server recovery > 8 hours 55 mins 800%
![Page 12: Answering the Database Scale Out Problem: SSDs in the Data Center April 14, 2010](https://reader038.fdocuments.us/reader038/viewer/2022110104/56815dd8550346895dcc0333/html5/thumbnails/12.jpg)
Typical DB read cluster
12
![Page 13: Answering the Database Scale Out Problem: SSDs in the Data Center April 14, 2010](https://reader038.fdocuments.us/reader038/viewer/2022110104/56815dd8550346895dcc0333/html5/thumbnails/13.jpg)
Fusion-io Value Add for Answers.com
• Scalability. >Twice the performance capacity on 1/4 servers
• 100% ROI on day of purchase (repurpose other 3/4)
• 75% reduction in operating costs: Rack Space, Power and Cooling Server Administration Database Administration
• 75% fewer failure points13
![Page 14: Answering the Database Scale Out Problem: SSDs in the Data Center April 14, 2010](https://reader038.fdocuments.us/reader038/viewer/2022110104/56815dd8550346895dcc0333/html5/thumbnails/14.jpg)
Other SSD uses in the Data Center
• Varnish (Web caching layer)• DB Backup Servers• Log Analysis• Data Warehouse
14
![Page 15: Answering the Database Scale Out Problem: SSDs in the Data Center April 14, 2010](https://reader038.fdocuments.us/reader038/viewer/2022110104/56815dd8550346895dcc0333/html5/thumbnails/15.jpg)
15
Thank you.
Slideshow: http://tinyurl.com/mysqlconf2010-ssd-dan