G
o
o
G
l
e
'
s
s
e
c
r
e
t
o
fly
s
U
c
c
e
s
s
?
2005-03-0908: 27- [Hand]
??????? March 3 Eclipsecon, Urs Hoelzle, Google Vice President of Engineering and Operations, I have a small secret of Google - how their data center is built - many people It is believed that Google's technology is the same as their search technology.
?????? In the history of Google's young operation, it is developed with the search algorithm to the information infrastructure management method. Google uses a huge amount of cheap goods, which uses effective management means to make them play a role that usually provides only high-end expensive large hosts. These unacceptable individuals support people from all the earth to Google Service request. ??????? The key is a grading and incorrect isolation. Hoelzle said that in the optimism, a household PC may be Crash once within 3 years. According to the size of Google, if there are thousands of PCs, there will be at least once a day. Therefore, Google will naturally use automated ways to deal with this situation - there is always something to do with the most intelligent brains - they have written a variety of tools to complete this task: ??????? Google Your own file system, Google File System, is designed to handle large-scale, optimized with 64M block unit. In addition, the GFS is designed to assume that the error will occur at any time. Google's data will be copied for 3 copies. If the original data is issued, a master machine can determine the copy of each data. With this fault-tolerant background support, it is possible to use the cheap PC to do Internet services. ??????? ??????? and flow-only, HOELZLE said that Google's PC server runs the cropped Linux, the redhat-based distribution - but - but the kernel is revised by Google .
??????? Google also designed a system for processing a large amount of data and a rapid response to the request. They divided the web into millions of Piece - or speaking Shard with Google Terminology. These shards will be replicated for fault tolerance. ??????? ??????? and everyone already understanded, Google creates index of the words on the Web that appears, and these indexes are used as large files of the series. Google certainly has a document server that stores a copy of the page crawling and downloaded. Another important thing is that Google is so directly when the distributed program that is written across thousands of machines is also ... Usually, build those servers in parallelism. Both applications require complex tools and techniques. Google's programming tools, MapReduce, you can also automatically restore an error program. Finally, Google also wrote "Batch" Job Scheduling Software Global Work Queue, as "supervisor", which divides the calculation job into multiple smaller tasks, distribute it to different computers.
??????? Some advanced technology support, Google gains relatively low operation overhead. Obviously Hoelzle's speech will not disclose any details of Google technology - even if you listen to the mouth ...
??????? news.com
