Oracle12.2之后，Dataguard推出了一个新的功能，Far sync Instance， 也就是 该实例不存放数据文件，只负责把相关在线日志数据或者归档数据同步到其他dg备端上，起到一个中转站的作用。具体可参考其他文章。 如本博文章： http://blog.itpub.net/29487349/viewspace-2792878/
下面的文章是转自mos，供有兴趣的小哥翻阅。
远程同步实例相关技术可参考: https://www.oracle.com/technetwork/database/availability/farsync-2267608.pdf
远同步实例的行为与为同步传输配置的任何其他数据保护/活动数据保护归档目标一样。这意味着关于影响往返网络延迟的光速的物理定律没有以任何方式改变；主数据库性能仍将受到主数据库和远同步实例之间的网络往返时间的影响。但是，通过将远同步实例放置在距离主数据库较远的metro距离，并允许远同步实例处理与数百或数千英里远的远程备用设备的所有通信，实现了实现远距离零数据丢失的灵活选项。这使所有情况下都能实现零数据丢失故障切换，但影响广泛地理区域的灾难除外，该灾难会同时导致主同步实例和远同步实例中断。很少有必要的配置最佳实践。大多数与适用于任何同步重做传输目标的相同。它们如下所示：
»主数据库和远程同步实例之间的网络必须满足 :
- Have round trip latency low enough so that the impact to response time and throughput of the primary database do es not exceed business requirements . The degree of impact is very application specific and wil l require testing to validate. In general, experience shows that there is a higher likelihood of success if the round - trip latency is less than 5ms, though there are successful deployments at higher latencies.
- Provide enough bandwidth between the primary database and far sync instance to accommodate peak redo volumes in addition to any other traffic sharing the network. Redo transport compression can be used to reduce network bandwidth requirements .
- Ideally there are redundant network links that will also tolerate network component failure.
»标准数据保护网络最佳实践，例如TCP发送和接收缓冲区大小的适当设置等于带宽延迟乘积的三倍。有关更多信息，请参阅数据保护网络传输的MAA实践。
(
http://docs.oracle.com/cd/E11882_01/server.112/e10803/config_dg.htm#HABPT5284)
»远同步实例的备用重做日志（SRL）应放置在具有足够IOPS（每秒写入数）容量的存储器上，以在峰值活动期间超过主数据库上LGWR进程的I/O of，以及其他活动的任何IOPS。这是一个重要的考虑因素。例如 :
- If the Far Sync instance has lesser performing disks than the primary it may not be able to forward redo to remote destinations as fast as it is received and a n archive log gap can form.
- In the case of redo gap resolution scenarios due to planned maintenance on the standby or network outages for example, there will be additional IO requests for gap resolution on top of redo from current transactions .
- Lesser performing disks at the Far Sync instance will delay acknowledgement to the primary database, increasing the total round - trip time between primary and standby and impacting application response time. This impact can be eliminated by u sing Fast Sync between the primary and the Far Sync instance .
»根据标准MAA最佳实践，远端同步实例的备用重做日志（SRL）的重做日志组数量应与主服务器上的相同，每个线程的重做日志组数量应为+1。 .
»备用远同步的SRL应在使用前手动清除，以便在激活备用远同步时实现最快的返回同步传输和零数据丢失保护。
»性能测试表明，小型远同步实例SGA不会影响远同步实例或主数据库的性能。MAA建议配置远同步功能所需的最小SGA。
»使用实例限制(http://docs.oracle.com/database/121/ADMIN/dbrm.htm#ADMIN13333)以实现尽可能最小的SGA。在测试期间减少CPU_计数对远同步实例的性能没有影响。
»MAA测试确定，在Linux上CPU_计数为1的300MB SGA 300足以进行远同步。（CPU_计数为1 SGA_目标为300M）
» 将远同步实例上的RMAN archivelog删除策略配置为“发送到备用”或“在备用时应用”，以自动管理远同步实例上的磁盘空间。如果主或备用实例有适当的备份计划，则无需备份远同步实例上的archivelog。
»为主数据库和备用数据库配置远同步实例，以便在角色转换后保持零数据丢失保护。部署在备用数据库城域距离内的远同步实例处于空闲状态，直到备用数据库成为主数据库
-
Note that
in
a
Data Guard Broker configuration a
switchover (planned role tr
ansition) cannot occur while in
Maximum Availability unless the
protection mode
can be enforced from the targe
t standby site.
If the standby
does not have its own Far Sync instance it will have to be configured to ship ASYNC to the original primar
y
after roles are reversed. T
his will prevent a switchover from occurring unless the pro
tection mode for the
primary database is first dropped from Maximum Availability to Maximum Performance.
»快速同步是Data Guard 12c的一项新功能，与4%到12%的同步相比，它提高了主数据库性能，具体取决于网络延迟和远同步实例硬件的I/O速度
»如果满足CPU、I/O和网络要求，将远同步实例放置在虚拟机上不会产生与物理硬件配置不同的性能
原文参考： Far Sync Configuration Best Practices (Doc ID 2678239.1)