！！rac的节点数实际中最多可以多少？！

flyhu · 发表于 2008-10-26 14:43

原帖由 stableman 于 2008-10-26 09:49 发表
我建议楼主用11.1.0.7 Linux平台来做RAC，这是一个非常稳定版本（可能先前9.1.xxx 10.1.xxx是不稳定版本给太多的人造成了xx.1.xxx就是不稳定版本的错觉）

如果希望自动均衡，只需要打开WRT这项功能(11.1.0.6/7), RAC DB能够根据每个节点的负载情况自动将负载均匀分布于各个节点

12个节点，建议用11个VD，两个OCR，private network的网络最好能用上1G, 建一个数据库就okay了，不需要太复杂

目前，RAC在理论上支持256个节点，在日本的实验室有一个128个节点的集群

谢谢哦，意见很具体，很实用，我们再具体实施的时候，回详细参考。

flyhu · 发表于 2008-10-26 15:06

原帖由 Yong Huang 于 2008-10-26 12:21 发表

Why don't you use server side loading balance (create a service and modify it with dbms_service)? I think Oracle recommends that over client side load balance.

I don't know why you got ORA-12170. Did you get any other errors in alert.log, udump, or listener log? Did you try SQL*Net trace to see if there's any clue?

By "cpu,内存，io并没有都拉起来", you mean those numbers stay low, i.e., those resources are not fully utilized. Correct? Server side load balance is controlled by the listeners, which have the latest node CPU usage (periodically updated from each node's PMON), so server side load balance may solve your uneven distribution problem.

"8节点的rac共享一个实例"? I don't understand. We have an 8-node RAC, each node running one instance. All access the same database. "实例" in Chinese means instance, not database.

Yong Huang

----------------------------------------------------------------------------
首先十分感谢，给我的回答很详细。我是一个系统架构师，还不是专业的ＤＢＡ，所以有些描述有些不够准确，见谅。
1、我们在应用服务器上的客户端tns是如下配的，不知道能否这样是否有问题，您说的server side loading balance，不知道是如何来做实施的？请多赐教。
NPSNSI =
  (DESCRIPTION =
(ADDRESS = (PROTOCOL = TCP)(HOST = npsnsi_dbs1-vip)(PORT = 1521))
(ADDRESS = (PROTOCOL = TCP)(HOST = npsnsi_dbs2-vip)(PORT = 1521))
(ADDRESS = (PROTOCOL = TCP)(HOST = npsnsi_dbs3-vip)(PORT = 1521))
(ADDRESS = (PROTOCOL = TCP)(HOST = npsnsi_dbs4-vip)(PORT = 1521))
(LOAD_BALANCE = yes)
(CONNECT_DATA =
   (SERVER = DEDICATED)
   (SERVICE_NAME = npsnsi)
   ( FAILOVER_MODE =
      (TYPE = SELECT)
      (METHOD = BASIC)
      (RETRIES = 10)
      (DELAY = 3)
   )
)
  )

2、我说的12170是应用端报的错，数据端我们的工程师说有alter中有600号错误。我们数据库原来运行的很好，只是过了10月份业务量有所上升，才出现这个问题，但我们监控四台数据库服务器并不是很忙。目前我们的数据库工程师正在查找原因。我们现在应用端临时采取了尝试重联的处理。

3、实例的，是我描述的错误，误导了您，不好意思。

4、我们即将实现的应用，是一个经营分析系统，主要操作是大量的统计及查询、基础数据是通过批量加载来实现的，预计未来3年数据在10-15T之间。
像这样应用，是不是节点间要同步的信息不会很大？！
我这边初步的想法：
方案一：是12个节点共用一个数据库，每个节点一个实例：2个节点夜间做数据加载，白天提供数据查询； 4个节点主要做统计分析，兼做一定的查询，4个节点做负载均衡；另外6个节点主要提供查询服务，6个节点做负载均衡。
想请教：在oracle rac中，是否可以想我描述这样，很灵活的将不同实例进行分组负载均衡。像我如上描述分成三组，每组内部做负载均衡。

方案二：将12个节点分成三组（每四台一组），建三个四节点rac数据库，每个rac数据库负责不同地域的统计分析支持。每个rac内部，一台用作数据加载，其他用作统计分析和查询；对于全国范围的分析，通过应用服务器来做聚合，将三个数据库的不同地域的结果合并在一起展现。

以上两个方案，不知道，大家有什么意见和建议。谢谢。

stableman · 发表于 2008-10-26 20:08

1> 您说的server side loading balance
在数据库服务器的Listener里面配置， Server端的Listener会pick up income connection, and forward to a node randomly.

2>不是很了解你的需求不便发表意见

flyhu · 发表于 2008-10-27 13:08

原帖由 stableman 于 2008-10-26 20:08 发表
1> 您说的server side loading balance
在数据库服务器的Listener里面配置， Server端的Listener会pick up income connection, and forward to a node randomly.

2>不是很了解你的需求不便发表意见

可以一同多讨论，想听听大家的建议与意见。

Yong Huang · 发表于 2008-10-28 00:51

原帖由 flyhu 于 2008-10-26 01:06 发表
1、我们在应用服务器上的客户端tns是如下配的，不知道能否这样是否有问题，您说的server side loading balance，不知道是如何来做实施的？请多赐教。
NPSNSI =
  (DESCRIPTION =
(ADDRESS = (PROTOCOL = TCP)(HOST = npsnsi_dbs1-vip)(PORT = 1521))
(ADDRESS = (PROTOCOL = TCP)(HOST = npsnsi_dbs2-vip)(PORT = 1521))
(ADDRESS = (PROTOCOL = TCP)(HOST = npsnsi_dbs3-vip)(PORT = 1521))
(ADDRESS = (PROTOCOL = TCP)(HOST = npsnsi_dbs4-vip)(PORT = 1521))
(LOAD_BALANCE = yes)
(CONNECT_DATA =
   (SERVER = DEDICATED)
   (SERVICE_NAME = npsnsi)
   ( FAILOVER_MODE =
      (TYPE = SELECT)
      (METHOD = BASIC)
      (RETRIES = 10)
      (DELAY = 3)
   )
)
  )
...
4、我们即将实现的应用，是一个经营分析系统，主要操作是大量的统计及查询、基础数据是通过批量加载来实现的，预计未来3年数据在10-15T之间。
...

Your TNS entry looks good. But make sure your service is configured correctly on the server side. You can check with srvctl and query v$services:

$ srvctl status service -d dbname -s servicename
Service servicename is running on instance(s) dbinstance1, dbinstance2, dbinstance3

SQL> select goal, clb_goal from v$services where name = 'servicename';

GOAL CLB_G
------------ -----
SERVICE_TIME LONG

We create services like this:
http://yong321.freeshell.org/oranotes/Service.txt

A good source about services, other than official documentation, is Jeremy Schneider's article:
http://www.ardentperf.com/pub/services-schneider07.pdf

Load balancing is explained in section 3.6.

As to your data warehouse design, you may refer to Bohan Chen's article, "Oracle RAC Large Scale Data Warehousing at Yahoo! Webseminar Slides", at
http://www.oracleracsig.org

(Free registration is needed.)

Yong Huang

flyhu · 发表于 2008-10-28 22:23

原帖由 Yong Huang 于 2008-10-28 00:51 发表

Your TNS entry looks good. But make sure your service is configured correctly on the server side. You can check with srvctl and query v$services:

$ srvctl status service -d dbname -s servicename
Service servicename is running on instance(s) dbinstance1, dbinstance2, dbinstance3

SQL> select goal, clb_goal from v$services where name = 'servicename';

GOAL CLB_G
------------ -----
SERVICE_TIME LONG

We create services like this:
http://yong321.freeshell.org/oranotes/Service.txt

A good source about services, other than official documentation, is Jeremy Schneider's article:
http://www.ardentperf.com/pub/services-schneider07.pdf

Load balancing is explained in section 3.6.

As to your data warehouse design, you may refer to Bohan Chen's article, "Oracle RAC Large Scale Data Warehousing at Yahoo! Webseminar Slides", at
http://www.oracleracsig.org

(Free registration is needed.)

Yong Huang

非常感谢给以的建议和提供的资料，我仔细去学习一下。thanks.

blue_prince · 发表于 2008-10-29 15:26

整过12个节点的RAC，现在运行好好的，不过由于应用复杂偶而会碰到BUG，其他还行。

flyhu · 发表于 2008-10-29 22:26

原帖由 blue_prince 于 2008-10-29 15:26 发表
整过12个节点的RAC，现在运行好好的，不过由于应用复杂偶而会碰到BUG，其他还行。

谢谢，非常好，终于有一个实际例子，可以为我增强信心了。

shahand · 发表于 2008-12-1 16:03

When using solely Oracle Clusterware: 63 nodes (9i or 10gR1)
With 10g Release 2, the maximum nodes is 100

When using a third party clusterware:

Sun: 8

HP UX: 16

HP Tru64: 8

IBM AIX:

* 8 nodes for Physical Shared (CLVM) SSA disk

* 16 nodes for Physical Shared (CLVM) non-SSA disk

* 128 nodes for Virtual Shared Disk (VSD)

* 128 nodes for GPFS

* Subject to storage subsystem limitations

Veritas: 8-16 nodes (check w/ Veritas)

Yong Huang · 发表于 2008-12-2 02:50

原帖由 shahand 于 2008-12-1 02:03 发表
When using solely Oracle Clusterware: 63 nodes (9i or 10gR1)
With 10g Release 2, the maximum nodes is 100
...

Where did you get this information? Or you mean that's how you found out by your own test?

Yong Huang

！！rac的节点数实际中最多可以多少？！

浏览过的版块