r*******n
发帖数: 3020 | |
w**z
发帖数: 8232 | 2 google 一下。
【在 r*******n 的大作中提到】
: re
|
c****e
发帖数: 1453 | 3 Depends your application. It generally holds more data, because it's simpler
. On the
other side, it has many limitations: the stupid db level lock blocks
reading while writing, no row larger than 64M, too many processes to config
sharding...
Do you know the joke "MongoDB is web scale"? Enjoy
http://www.youtube.com/watch?v=b2F-DItXtZs |
r*******n
发帖数: 3020 | 4 google了，自己学识有限，依然不清晰
【在 w**z 的大作中提到】
: google 一下。
|
t********e
发帖数: 880 | 5 mongodb就一坨屎，性能一塌糊涂，查询能力极弱 |
N*****m
发帖数: 42603 | 6 先说说你的应用，才能谈到有什么优势
【在 r*******n 的大作中提到】
: google了，自己学识有限，依然不清晰
|
r*******n
发帖数: 3020 | 7 现在我们用SQL server；看到mongoDB，就像了解下，
做什么有优势，我们可以采用。
我们数据库存了大量的从设备采集来的数据， 数据结构类似：
{timestamp， value, name}
【在 N*****m 的大作中提到】
: 先说说你的应用，才能谈到有什么优势
|
w**z
发帖数: 8232 | 8 Mongo stores Bson , requires whole dataset in memory in order to perform.
建议你看看nosql in general. two good paper to read, Amazon dynamo and
Google big table.
【在 r*******n 的大作中提到】
: google了，自己学识有限，依然不清晰
|
w**z
发帖数: 8232 | 9 take a look at Cassandra and hbase also . is your app read or write heavy
? data size? work load? consistency requirement ? HA requirement? nosql
doesn't fit all the use cases , rdbms serves majority of the use cases just
fine
【在 r*******n 的大作中提到】
: 现在我们用SQL server；看到mongoDB，就像了解下，
: 做什么有优势，我们可以采用。
: 我们数据库存了大量的从设备采集来的数据， 数据结构类似：
: {timestamp， value, name}
|
N*****m
发帖数: 42603 | 10 如果是write once, read many times，用mongodb不错
【在 r*******n 的大作中提到】
: 现在我们用SQL server；看到mongoDB，就像了解下，
: 做什么有优势，我们可以采用。
: 我们数据库存了大量的从设备采集来的数据， 数据结构类似：
: {timestamp， value, name}
|
|
|
r*******n
发帖数: 3020 | 11 The data comes in the warehouse every a minute 24/7 and will be used for
data analysis.
it's kind of fit "write once, read many"
heavy
just
【在 w**z 的大作中提到】
: take a look at Cassandra and hbase also . is your app read or write heavy
: ? data size? work load? consistency requirement ? HA requirement? nosql
: doesn't fit all the use cases , rdbms serves majority of the use cases just
: fine
|
w**z
发帖数: 8232 | 12 what kind of Query are you going to use? What's wrong with the mysql? nosql
is not good for analytics in general.
【在 r*******n 的大作中提到】
: The data comes in the warehouse every a minute 24/7 and will be used for
: data analysis.
: it's kind of fit "write once, read many"
:
: heavy
: just
|
g*****g
发帖数: 34805 | 13 If you don't have much relational data, you may consider Cassandra instead.
NoSQL is harder than SQL DB. You only need NoSQL when you are trying to
achieve high availability or reduce licensing cost.
【在 r*******n 的大作中提到】
: 现在我们用SQL server；看到mongoDB，就像了解下，
: 做什么有优势，我们可以采用。
: 我们数据库存了大量的从设备采集来的数据， 数据结构类似：
: {timestamp， value, name}
|
r*******n
发帖数: 3020 | 14 设备会很多，多数是sensor，数据量会比较大并且同时进来。
如果mongoDB吞吐能力强并且占用硬盘空间小，我们可以把它作为中转站，再二次处理
放到SQL server里供分析用，
同时也可作为数据备份用。
MySql肯定能做， 想看看mongoDB是不是更合适。
nosql
【在 w**z 的大作中提到】
: what kind of Query are you going to use? What's wrong with the mysql? nosql
: is not good for analytics in general.
|
w**z
发帖数: 8232 | 15 if it has concurrent writes with high volume, Cassandra is better, the write
is sequential disk write, extremely fast (<1ms). For Mongo, need to read
before write since it's a Bson on disk.
You can use Hadoop integration with Cassandra to read data out. But don't
expect much for relational flavor of sql from Cassandra. Think of it as a
big Hashtable, no join, very poor index support. It's good at HA and large
volume writes.
【在 r*******n 的大作中提到】
: 设备会很多，多数是sensor，数据量会比较大并且同时进来。
: 如果mongoDB吞吐能力强并且占用硬盘空间小，我们可以把它作为中转站，再二次处理
: 放到SQL server里供分析用，
: 同时也可作为数据备份用。
: MySql肯定能做， 想看看mongoDB是不是更合适。
:
: nosql
|
g*****g
发帖数: 34805 | 16 Cassandra read is also fast and tunable. The key is the data relation.
It's a hashtable of hashtable and it's best used as a hash table though
relation can be achieved, with labor.
write
【在 w**z 的大作中提到】
: if it has concurrent writes with high volume, Cassandra is better, the write
: is sequential disk write, extremely fast (<1ms). For Mongo, need to read
: before write since it's a Bson on disk.
: You can use Hadoop integration with Cassandra to read data out. But don't
: expect much for relational flavor of sql from Cassandra. Think of it as a
: big Hashtable, no join, very poor index support. It's good at HA and large
: volume writes.
|
w**z
发帖数: 8232 | 17 Sure, Cassandra's read is fast also, but it might involve multiple disk
accesses based on the number of SSTables it needs to read from. So for us it
's the average is 10ms vs 1ms :)
Do you have any experience with HBase? From the benchmark, it's even better
than Cassandra, but it's a lot more complicated to manage.
By the way, there is Cassandra Summit next week in SF and Hbase conference
the day after. I am going to both of them. Are you going by any chance?
【在 g*****g 的大作中提到】
: Cassandra read is also fast and tunable. The key is the data relation.
: It's a hashtable of hashtable and it's best used as a hash table though
: relation can be achieved, with labor.
:
: write
|
s********k
发帖数: 6180 | 18 做数据中心的energy management？肯定是data import和analysis分开，应该会有专门
的local sever和cloud，local server只做暂存和data import，肯定不属于于一次写
多次读吧，搞不好写的次数比读的次数还多（如果local做了很多处理，传到cloud的数
据量应该会小一些），cloud还是得用SQL了。
我们做的产品很像，我们是local用sqllite+SQAacademy， cloud做的是postgreSQL。
SQL只要做好cache其实性能不差的，你们的应用应该是数据量大，但是可能burst不明
显的，这样比较规律的数据优化好SQL肯定可以搞。
【在 r*******n 的大作中提到】
: 设备会很多，多数是sensor，数据量会比较大并且同时进来。
: 如果mongoDB吞吐能力强并且占用硬盘空间小，我们可以把它作为中转站，再二次处理
: 放到SQL server里供分析用，
: 同时也可作为数据备份用。
: MySql肯定能做， 想看看mongoDB是不是更合适。
:
: nosql
|
g*****g
发帖数: 34805 | 19 No, I am not going. I should have registered.
it
better
【在 w**z 的大作中提到】
: Sure, Cassandra's read is fast also, but it might involve multiple disk
: accesses based on the number of SSTables it needs to read from. So for us it
: 's the average is 10ms vs 1ms :)
: Do you have any experience with HBase? From the benchmark, it's even better
: than Cassandra, but it's a lot more complicated to manage.
: By the way, there is Cassandra Summit next week in SF and Hbase conference
: the day after. I am going to both of them. Are you going by any chance?
|
c****e
发帖数: 1453 | 20 If you have continous writing, be careful that reading could be blocked on
small writes in MongoDB.
【在 r*******n 的大作中提到】
: 现在我们用SQL server；看到mongoDB，就像了解下，
: 做什么有优势，我们可以采用。
: 我们数据库存了大量的从设备采集来的数据， 数据结构类似：
: {timestamp， value, name}
|
|
|
c****e
发帖数: 1453 | 21 In his case, it seems append only, Cassandra not necessarily performance
better than MongoDB. Writing to memory mapped file is really fast.
write
【在 w**z 的大作中提到】
: if it has concurrent writes with high volume, Cassandra is better, the write
: is sequential disk write, extremely fast (<1ms). For Mongo, need to read
: before write since it's a Bson on disk.
: You can use Hadoop integration with Cassandra to read data out. But don't
: expect much for relational flavor of sql from Cassandra. Think of it as a
: big Hashtable, no join, very poor index support. It's good at HA and large
: volume writes.
|
r*******n
发帖数: 3020 | 22 我们做的差不多，thank you for sharing.
【在 s********k 的大作中提到】
: 做数据中心的energy management？肯定是data import和analysis分开，应该会有专门
: 的local sever和cloud，local server只做暂存和data import，肯定不属于于一次写
: 多次读吧，搞不好写的次数比读的次数还多（如果local做了很多处理，传到cloud的数
: 据量应该会小一些），cloud还是得用SQL了。
: 我们做的产品很像，我们是local用sqllite+SQAacademy， cloud做的是postgreSQL。
: SQL只要做好cache其实性能不差的，你们的应用应该是数据量大，但是可能burst不明
: 显的，这样比较规律的数据优化好SQL肯定可以搞。
|
a***n
发帖数: 538 | |
z*******3
发帖数: 13709 | 24 用nosql就是为了摆脱transaction这些在并发量上去时候产生的资源瓶颈
mongodb将所有index全部读入内存，等于是刚出虎口又入狼穴
【在 c****e 的大作中提到】
: In his case, it seems append only, Cassandra not necessarily performance
: better than MongoDB. Writing to memory mapped file is really fast.
:
: write
|
w***g
发帖数: 5958 | 25 同意.
【在 t********e 的大作中提到】
: mongodb就一坨屎，性能一塌糊涂，查询能力极弱
|
s*****n
发帖数: 5488 | 26 Foursquare uses MongoDB. So I would say performance is not an issue.
mongoDB的数据最后悔比mysql 大。可能又3倍以上。foursquare用两台70G的server就
搞定了。
用不用mongoDB还要看你们的需要。例如数据是不是schemaless.
比如timestamp, sensor, value, tag格式。
如果说value或者需求会经常变化，以location为例，
可以是gps location or wifi
一个是
lan/lat, accuracy,bearing, speed
一个只有ll+ accc
或者有将来扩展的需求，例如bluetooth location
或者是经常被tag,tag一些trace, debug info.
例如tag calling locID = {stream}
这样用mongoDB会很灵活。
否则的话，mysql很好用了。
【在 r*******n 的大作中提到】
: The data comes in the warehouse every a minute 24/7 and will be used for
: data analysis.
: it's kind of fit "write once, read many"
:
: heavy
: just
|
