w**z 发帖数: 8232 | 1 Netflix has almost everything running with Cassandra on AWS. it is pretty
amazing how they do things. It was pretty eye opening how Netflix trades
data consistency for HA and low latency. They use CL.One for good portion of
the cassandra operations. When I really think about it, it does make a lot
of sense. We can compensate data in-consistency with client side
contingency plan. In most of the cases, we don't really need the strong
consistency RDBMS system can offer. It's all about BIG data nowadays. |
z****e 发帖数: 54598 | 2 这对开发来说是好事
mapreduce和nosql都需要码农去搞
db则多数时候dba就可以了 |
c*********e 发帖数: 16335 | 3 mapreduce就是一hashtable.
【在 z****e 的大作中提到】 : 这对开发来说是好事 : mapreduce和nosql都需要码农去搞 : db则多数时候dba就可以了
|
c****e 发帖数: 1453 | 4 That's because almost all use scenarios in Netflix don't care about
consistency that much. I bet they won't run the billing pipeline out of
RDBMS. Most likely, they still have to pay premium to Oracle to charge money
in a reliable way.
There are many aspects of big data. It just depends on your application. You
can even have MPP RDBMS. |
c****e 发帖数: 1453 | 5 mapreduce has nothing to do with hashtable. I guess you want to say
Cassandra is a big hash table with column families.
【在 c*********e 的大作中提到】 : mapreduce就是一hashtable.
|
z*******3 发帖数: 13709 | 6 mapreduce和nosql只是两个概念
如何impl有很多种方式
前者完全可以用j2ee标准组件来做,spring正在进攻这个领域
1.0版本已经出来了
nosql作为db的补充,跟db配合可以对付绝大多数需求
剩下的部分交给主机去做
【在 c*********e 的大作中提到】 : mapreduce就是一hashtable.
|
g*****g 发帖数: 34805 | 7 It's transitioning to RDS on Amazon. Billing isn't exactly a high frequency
operation in Netflix. The challenge in Netflix is running at this scale in
the cloud, not the scale itself .
money
You
【在 c****e 的大作中提到】 : That's because almost all use scenarios in Netflix don't care about : consistency that much. I bet they won't run the billing pipeline out of : RDBMS. Most likely, they still have to pay premium to Oracle to charge money : in a reliable way. : There are many aspects of big data. It just depends on your application. You : can even have MPP RDBMS.
|
h***i 发帖数: 1970 | 8 也是,如果都是watch history, recommendation,favorite list的话,的确不需要什么
consistency, latency和availability更重要.
frequency
【在 g*****g 的大作中提到】 : It's transitioning to RDS on Amazon. Billing isn't exactly a high frequency : operation in Netflix. The challenge in Netflix is running at this scale in : the cloud, not the scale itself . : : money : You
|
h***i 发帖数: 1970 | 9 如果replication factor是3, consistency level 是quorum,latency是大一些,但是也
不慢吧? 这样可以strong consistency.
frequency
【在 g*****g 的大作中提到】 : It's transitioning to RDS on Amazon. Billing isn't exactly a high frequency : operation in Netflix. The challenge in Netflix is running at this scale in : the cloud, not the scale itself . : : money : You
|
w**z 发帖数: 8232 | 10 according to netflix, quorum is 3 times slower for rf =3,for their use case
which is a bit hard to believe
. so they use cl.one for some of the use cases.
【在 h***i 的大作中提到】 : 如果replication factor是3, consistency level 是quorum,latency是大一些,但是也 : 不慢吧? 这样可以strong consistency. : : frequency
|
w**z 发帖数: 8232 | 11 so Netflix is going to host everything on Amazon?
frequency
【在 g*****g 的大作中提到】 : It's transitioning to RDS on Amazon. Billing isn't exactly a high frequency : operation in Netflix. The challenge in Netflix is running at this scale in : the cloud, not the scale itself . : : money : You
|
g*****g 发帖数: 34805 | 12 Yes, it's 95% on AWS already, and it will be 100% by EOY.
【在 w**z 的大作中提到】 : so Netflix is going to host everything on Amazon? : : frequency
|
c****e 发帖数: 1453 | 13 Right, I just realized due to the subscription model payment transactions
could be well scheduled for Netflix and won't be on the blocking path,
compared with other online vendors, let's say Amazon.
frequency
【在 g*****g 的大作中提到】 : It's transitioning to RDS on Amazon. Billing isn't exactly a high frequency : operation in Netflix. The challenge in Netflix is running at this scale in : the cloud, not the scale itself . : : money : You
|