M*d
2 楼
【 以下文字转载自 Programming 讨论区 】
发信人: MoonChild (df), 信区: Programming
标 题: 一般distributed system用什么consistency model
发信站: BBS 未名空间站 (Tue Apr 18 17:22:40 2017, 美东)
有两个server, 互为replica, 每个都和自己的backend talk, 但database 是同一个。
要求是要implement consistency+in-memory cache, 具体哪种consistency 也不知道
。还要fault-tolerance
有什么思路呢多谢
发信人: MoonChild (df), 信区: Programming
标 题: 一般distributed system用什么consistency model
发信站: BBS 未名空间站 (Tue Apr 18 17:22:40 2017, 美东)
有两个server, 互为replica, 每个都和自己的backend talk, 但database 是同一个。
要求是要implement consistency+in-memory cache, 具体哪种consistency 也不知道
。还要fault-tolerance
有什么思路呢多谢
g*4
3 楼
啊,人呢?
真是回荡绝顶啊!
真是回荡绝顶啊!
d*1
4 楼
包子
能pm一下是啥东西么
能pm一下是啥东西么
j*r
5 楼
两个同样前端结点,两个同样后端节点。数据库加个standby replica就差不多了。
【在 M*******d 的大作中提到】![](/moin_static193/solenoid/img/up.png)
: 【 以下文字转载自 Programming 讨论区 】
: 发信人: MoonChild (df), 信区: Programming
: 标 题: 一般distributed system用什么consistency model
: 发信站: BBS 未名空间站 (Tue Apr 18 17:22:40 2017, 美东)
: 有两个server, 互为replica, 每个都和自己的backend talk, 但database 是同一个。
: 要求是要implement consistency+in-memory cache, 具体哪种consistency 也不知道
: 。还要fault-tolerance
: 有什么思路呢多谢
【在 M*******d 的大作中提到】
![](/moin_static193/solenoid/img/up.png)
: 【 以下文字转载自 Programming 讨论区 】
: 发信人: MoonChild (df), 信区: Programming
: 标 题: 一般distributed system用什么consistency model
: 发信站: BBS 未名空间站 (Tue Apr 18 17:22:40 2017, 美东)
: 有两个server, 互为replica, 每个都和自己的backend talk, 但database 是同一个。
: 要求是要implement consistency+in-memory cache, 具体哪种consistency 也不知道
: 。还要fault-tolerance
: 有什么思路呢多谢
l*l
7 楼
pai
s*3
8 楼
既然同一个Db 因该没consistenct 的问题? 重点是如何fault tolerance?
【在 M*******d 的大作中提到】
![](/moin_static193/solenoid/img/up.png)
: 【 以下文字转载自 Programming 讨论区 】
: 发信人: MoonChild (df), 信区: Programming
: 标 题: 一般distributed system用什么consistency model
: 发信站: BBS 未名空间站 (Tue Apr 18 17:22:40 2017, 美东)
: 有两个server, 互为replica, 每个都和自己的backend talk, 但database 是同一个。
: 要求是要implement consistency+in-memory cache, 具体哪种consistency 也不知道
: 。还要fault-tolerance
: 有什么思路呢多谢
s*r
9 楼
co-han, hope we see it next week
k*i
14 楼
一般用 paxos
i*e
17 楼
你的问题不清楚。
你弄清:系统是用在什么地方,是否是地理上离很远,还是在同一个数据中心,是否两
个服务器是active active or active passive back up. Or whether there is a
primary replica and secondary replica, and replica management policy.
From request perspective, whether it's read intensive and/or write intensive
.
Once understood the above, then you can select consistency model/semantic.
1. Complete consistency. It means whenever a data item is updated, both
servers will be able to read this copy of data. In academic research, this
can be assumed. In reality, this is not achievable since it requests global
clock that does not exist in distributed systems.
2. Sequential consistency. This relaxed the requirement of global clock a
bit. It assumes predefined sequential order of operations. Different
operations may execute at different time intervals, but following the order.
If so, the consistency is guaranteed.
3. Linear consistency. Stringer than 2.
4. ...
5. Eventual consistency. This is widely used in production system. It
relaxed the time requirement to minutes. It means eventually different
copies of the data will be consistent. Pro. Easier to implement, for short
period of time different copies of the same data exist. Eventual consistency
is used by Azure, AWS, GCP etc.
For in memory cache,the key trade off is network overhead vs. freshness of
cache data.
Fault tolerance is much more complicated. Will address later if having time
later on.
你弄清:系统是用在什么地方,是否是地理上离很远,还是在同一个数据中心,是否两
个服务器是active active or active passive back up. Or whether there is a
primary replica and secondary replica, and replica management policy.
From request perspective, whether it's read intensive and/or write intensive
.
Once understood the above, then you can select consistency model/semantic.
1. Complete consistency. It means whenever a data item is updated, both
servers will be able to read this copy of data. In academic research, this
can be assumed. In reality, this is not achievable since it requests global
clock that does not exist in distributed systems.
2. Sequential consistency. This relaxed the requirement of global clock a
bit. It assumes predefined sequential order of operations. Different
operations may execute at different time intervals, but following the order.
If so, the consistency is guaranteed.
3. Linear consistency. Stringer than 2.
4. ...
5. Eventual consistency. This is widely used in production system. It
relaxed the time requirement to minutes. It means eventually different
copies of the data will be consistent. Pro. Easier to implement, for short
period of time different copies of the same data exist. Eventual consistency
is used by Azure, AWS, GCP etc.
For in memory cache,the key trade off is network overhead vs. freshness of
cache data.
Fault tolerance is much more complicated. Will address later if having time
later on.
m*l
18 楼
顶
M*d
22 楼
thanks a lot.
Actually it's a project for a distributed system course
I have no idea at all how to implement. The requirement is kinda vague as
well.
It just asks to implement all the concept.
The system is internet of things, where you have sensors/devices and a
gateway(controller) and a backend server(persistence layer)
I can think of a replica both for the gateway and backend, but I still have
to think about it.
intensive
global
【在 i*****e 的大作中提到】![](/moin_static193/solenoid/img/up.png)
: 你的问题不清楚。
: 你弄清:系统是用在什么地方,是否是地理上离很远,还是在同一个数据中心,是否两
: 个服务器是active active or active passive back up. Or whether there is a
: primary replica and secondary replica, and replica management policy.
: From request perspective, whether it's read intensive and/or write intensive
: .
: Once understood the above, then you can select consistency model/semantic.
: 1. Complete consistency. It means whenever a data item is updated, both
: servers will be able to read this copy of data. In academic research, this
: can be assumed. In reality, this is not achievable since it requests global
Actually it's a project for a distributed system course
I have no idea at all how to implement. The requirement is kinda vague as
well.
It just asks to implement all the concept.
The system is internet of things, where you have sensors/devices and a
gateway(controller) and a backend server(persistence layer)
I can think of a replica both for the gateway and backend, but I still have
to think about it.
intensive
global
【在 i*****e 的大作中提到】
![](/moin_static193/solenoid/img/up.png)
: 你的问题不清楚。
: 你弄清:系统是用在什么地方,是否是地理上离很远,还是在同一个数据中心,是否两
: 个服务器是active active or active passive back up. Or whether there is a
: primary replica and secondary replica, and replica management policy.
: From request perspective, whether it's read intensive and/or write intensive
: .
: Once understood the above, then you can select consistency model/semantic.
: 1. Complete consistency. It means whenever a data item is updated, both
: servers will be able to read this copy of data. In academic research, this
: can be assumed. In reality, this is not achievable since it requests global
L*s
23 楼
Receiving a Refund
We'll request a refund after we receive and process your return. If you are
due a refund for non-returnable items, such as groceries, we'll issue a refu
nd when you contact us.
For items directed to our Returns Center, expect your refund within 4 weeks
of giving your package to the return shipper (though in many cases you'll re
ceive the refund sooner).
This time period includes:
* 5 to 10 business days for us to receive your return from the shipper
* 3 to 5 business days for us to process your return
* The time it takes your bank to process our refund request (see below f
or specific guidelines)
We'll notify you via e-mail of your refund once we've received and processed
the returned item.
You can view completed returns and refunds at the bottom of an order's summa
ry page in Your Account. If the return and refund do not appear in Your Acco
unt within 4 weeks from the day you sent us your return, please e-mail us fo
r help.
【在 w***a 的大作中提到】![](/moin_static193/solenoid/img/up.png)
: 真以为送给你了?
We'll request a refund after we receive and process your return. If you are
due a refund for non-returnable items, such as groceries, we'll issue a refu
nd when you contact us.
For items directed to our Returns Center, expect your refund within 4 weeks
of giving your package to the return shipper (though in many cases you'll re
ceive the refund sooner).
This time period includes:
* 5 to 10 business days for us to receive your return from the shipper
* 3 to 5 business days for us to process your return
* The time it takes your bank to process our refund request (see below f
or specific guidelines)
We'll notify you via e-mail of your refund once we've received and processed
the returned item.
You can view completed returns and refunds at the bottom of an order's summa
ry page in Your Account. If the return and refund do not appear in Your Acco
unt within 4 weeks from the day you sent us your return, please e-mail us fo
r help.
【在 w***a 的大作中提到】
![](/moin_static193/solenoid/img/up.png)
: 真以为送给你了?
L*s
25 楼
这个才是需要return的流程吧
哪有还没收到return的东西就先退钱的...
are
refu
weeks
re
【在 L****s 的大作中提到】![](/moin_static193/solenoid/img/up.png)
: Receiving a Refund
: We'll request a refund after we receive and process your return. If you are
: due a refund for non-returnable items, such as groceries, we'll issue a refu
: nd when you contact us.
: For items directed to our Returns Center, expect your refund within 4 weeks
: of giving your package to the return shipper (though in many cases you'll re
: ceive the refund sooner).
: This time period includes:
: * 5 to 10 business days for us to receive your return from the shipper
: * 3 to 5 business days for us to process your return
哪有还没收到return的东西就先退钱的...
are
refu
weeks
re
【在 L****s 的大作中提到】
![](/moin_static193/solenoid/img/up.png)
: Receiving a Refund
: We'll request a refund after we receive and process your return. If you are
: due a refund for non-returnable items, such as groceries, we'll issue a refu
: nd when you contact us.
: For items directed to our Returns Center, expect your refund within 4 weeks
: of giving your package to the return shipper (though in many cases you'll re
: ceive the refund sooner).
: This time period includes:
: * 5 to 10 business days for us to receive your return from the shipper
: * 3 to 5 business days for us to process your return
j*n
26 楼
Cache 一定要in server的memory吗?redis行不行
c*e
27 楼
贪心贪到英文都不认识了?
are
refu
weeks
re
【在 L****s 的大作中提到】![](/moin_static193/solenoid/img/up.png)
: Receiving a Refund
: We'll request a refund after we receive and process your return. If you are
: due a refund for non-returnable items, such as groceries, we'll issue a refu
: nd when you contact us.
: For items directed to our Returns Center, expect your refund within 4 weeks
: of giving your package to the return shipper (though in many cases you'll re
: ceive the refund sooner).
: This time period includes:
: * 5 to 10 business days for us to receive your return from the shipper
: * 3 to 5 business days for us to process your return
are
refu
weeks
re
【在 L****s 的大作中提到】
![](/moin_static193/solenoid/img/up.png)
: Receiving a Refund
: We'll request a refund after we receive and process your return. If you are
: due a refund for non-returnable items, such as groceries, we'll issue a refu
: nd when you contact us.
: For items directed to our Returns Center, expect your refund within 4 weeks
: of giving your package to the return shipper (though in many cases you'll re
: ceive the refund sooner).
: This time period includes:
: * 5 to 10 business days for us to receive your return from the shipper
: * 3 to 5 business days for us to process your return
l*i
34 楼
你是不是打算把东西留下,然后白吃AMZ600. 呵呵
贪小便宜的人往往吃大亏
贪小便宜的人往往吃大亏
c*7
36 楼
发包子就可以了
D*g
37 楼
pai
b*6
39 楼
先排包子
相关阅读