a********r
发帖数: 218 | 1 How do you design (or implement) a scale-out distributed system to grant a
good performance?
I said multithreaded programming, sychronization technique. But, he want the
details of the implementation.
那位达人能给个线素？any pattern possible? | a*****g
发帖数: 110 | 2 narrow down the questions first, like what is it used for so you can figure
out what data structure is needed, and how to distribute data and workloads
the
【在 a********r 的大作中提到】
: How do you design (or implement) a scale-out distributed system to grant a
: good performance?
: I said multithreaded programming, sychronization technique. But, he want the
: details of the implementation.
: 那位达人能给个线素？any pattern possible?
| s*******f
发帖数: 1114 | 3 good performance means less data transfer. so in mapreduce, try to always
let the data be processed in where it located, or if must transfer data, try
to transfer within rack.
As you map missions to different machine, some may slower. The most slow
machine (caused by bad hard disk or heavy work, etc) will often the
bottleneck because other machine need to wait it to finish. Distribute the
missions in slow machine to other relative leisure machines can improve
performance.
For multi-thread, thread should belong to process, right? different machine
means different processes run |
|
