s***o
发帖数: 2191 | 1 http://www.sarahmei.com/blog/2013/11/11/why-you-should-never-us
It's a lengthy but very interesting article as it illustrates one of my
major concerns for using mongodb as primary data store. Any thoughts? |
z****e
发帖数: 54598 | 2 定义一下primary和secondary
操作快和数据新的放在primary还是secondary? |
z****e
发帖数: 54598 | 3 扫了一遍
他用的例子恰好是我回避过的
也就是用户数据，比如账户信息这种
还有涉及到金钱交易的数据
都建议用database，因为transaction和join是你非常需要的
否则打散了的数据会带来经济上巨大的损失
但是，对于一些非重要数据
比如评论，你可以把评论本身单独存成documents
然后添加table id in db
这样就把靠近系统的缓存一分为二
把重要的数据交给db，把非重要数据交给c*,mongodb or couchdb
或者随便你挑一个ap system就好了
然后因为你无法保证ap system里面数据是consistent的
所以你最后还需要一个cp system which usually is hbase
然后把hbase当data warehouse用
只存只读不改不删，酱紫，就可以回避掉这篇文章里面说的del所带来的问题
其实这都在内森那篇文章里写了
内森写得多好，可惜看的人不多 |
z****e
发帖数: 54598 | 4 该用db时候千万不要省
否则麻烦大大地有
这篇文章里面用户数据存成nosql那这就是一个大错
评论什么存在nosql里面无所谓了
错就错了，又怎样，买卖提的帖子数据就是documents |
z****e
发帖数: 54598 | 5 pure db
or
pure nosql
都是错误的
not only sql才是正确的
但是很明显
现在做成db超集的nosql并不多
所以最好的方式还是凑几个一起用
取长补短
我一般用postgresql, cassandra和hbase的组合 |
i********y
发帖数: 153 | 6 可以分享一下文章的链接吗，谢谢
★ 发自iPhone App: ChineseWeb 8.2.2
【在 z****e 的大作中提到】
: 扫了一遍
: 他用的例子恰好是我回避过的
: 也就是用户数据，比如账户信息这种
: 还有涉及到金钱交易的数据
: 都建议用database，因为transaction和join是你非常需要的
: 否则打散了的数据会带来经济上巨大的损失
: 但是，对于一些非重要数据
: 比如评论，你可以把评论本身单独存成documents
: 然后添加table id in db
: 这样就把靠近系统的缓存一分为二
|
p*****w
发帖数: 429 | 7 mongo适合数据migration, 适合数据结构经常变化的应用. 存用户之类的主数据库可能
就不如其他有优势.
【在 s***o 的大作中提到】
: http://www.sarahmei.com/blog/2013/11/11/why-you-should-never-us
: It's a lengthy but very interesting article as it illustrates one of my
: major concerns for using mongodb as primary data store. Any thoughts?
|
d****n
发帖数: 12461 | 8 我觉得你说的不是一回事。
【在 z****e 的大作中提到】
: 扫了一遍
: 他用的例子恰好是我回避过的
: 也就是用户数据，比如账户信息这种
: 还有涉及到金钱交易的数据
: 都建议用database，因为transaction和join是你非常需要的
: 否则打散了的数据会带来经济上巨大的损失
: 但是，对于一些非重要数据
: 比如评论，你可以把评论本身单独存成documents
: 然后添加table id in db
: 这样就把靠近系统的缓存一分为二
|
c******o
发帖数: 1277 | |
N********n
发帖数: 8363 | 10 Qutoe: "Schema flexibility sounds like a great idea, but the only time
it's actually useful is when the structure of your data has no value"
LOL. Like I have said for a trillion times already, the so called
"big data" is only good at handling data w/o a fixed schema also
known as JUNK. How much value can you mine outta junk?
If your data has any kinda importance then it naturally has a structure
or a form of relationship that needs to be organized into a schema. If
you think some schema-less "big data" snake oil solution could somehow
let you get away w/ it then you are just kidding yourself like the
quoted author has found out. |
g*****g
发帖数: 34805 | 11 90% of our data is in Cassandra/HBase, and billions of $ are generated out
of it. Don't speak of things you don't know.
Not every application is social app, in fact, users in most applications don
't have relationships with other users. MitBBS is an example itself.
【在 N********n 的大作中提到】
: Qutoe: "Schema flexibility sounds like a great idea, but the only time
: it's actually useful is when the structure of your data has no value"
: LOL. Like I have said for a trillion times already, the so called
: "big data" is only good at handling data w/o a fixed schema also
: known as JUNK. How much value can you mine outta junk?
: If your data has any kinda importance then it naturally has a structure
: or a form of relationship that needs to be organized into a schema. If
: you think some schema-less "big data" snake oil solution could somehow
: let you get away w/ it then you are just kidding yourself like the
: quoted author has found out.
|
N********n
发帖数: 8363 | 12
NFLX gets its income from the media they offer instead of some "big
data" they host unless you wanna stretch it to define all the media
contents as data. That's BULL, and you know it.
【在 g*****g 的大作中提到】
: 90% of our data is in Cassandra/HBase, and billions of $ are generated out
: of it. Don't speak of things you don't know.
: Not every application is social app, in fact, users in most applications don
: 't have relationships with other users. MitBBS is an example itself.
|
g*****g
发帖数: 34805 | 13 For engineering, it's how to serve the catalog, not the catalog itself that'
s interesting. 75% of the content are served through recommendation, not
search. The recommendation is from user data analysis.
Even for the catalog itself, what will be the big bang for the buck is from
big data analysis. All purchase decisions are based on the data.
If two stores are selling baozi, you can say the value is from baozi, but
what makes one successful and another bankruptcy is the skills making the
baozi.
【在 N********n 的大作中提到】
:
: NFLX gets its income from the media they offer instead of some "big
: data" they host unless you wanna stretch it to define all the media
: contents as data. That's BULL, and you know it.
|
