SVC的ESC拉伸双活模式Lun是否可以就近读写？hperswap模式是否可以就近读写？
SVC的ESC拉伸双活模式Lun是否可以就近读写？hperswap模式lun是否可以就近读写？还是只能读对端lun？造成这个问题的原因是什么？
收起查看其它 6 个回答jxnxsdengyu的回答
SVC StretchCluster与SVC HyperSwap的最大特性就是SVC节点“站点化”，主机节点“站点化”，存储节点“站点化”，所以这两种模式都是同一站点的主机读写同一站点的SVC的节点，SVC节点读写同一站点的存储节点。
所以这两种ACTIVE-ACTIVE存储双活方案，是就近读写的，另外SVC HyperSwap不仅仅是就近读写，如果本地站点的IO流量连续10分钟低于25%，而对端站点的IO流量连续10分钟高于75%，那么SVC HyperSwap将反转读写关系，优先读写对端站点。
- 感谢解答，但我认为hyperwap双活模式下，不可以就近读写。两个站点都要去找该lun的owner站点去写，然后镜像到另一端缓存。因SVC与vplex的全局一致缓存不同SVC的hperswap是两套缓存表。正式因为这种情况，导致会有主机频繁写对端lun的情况出现，导致lun owner关系反转。 不知道我的看法是否正确？谢谢
- 两端主机是各自访问各端的SVC，但AUX端在未切换为master时是不可被主机读写的。AUX端主机的IO是要同端的SVC 转发到Master端的SVC的。红皮书原文： The Primary site processes all reads and writes, so reads to the non-Primary site have an increase in latency (of 1x the round-trip time (RTT) between the sites) while the data is retrieved over the Private SAN. This increase in latency causes an increase in Primary-to-non-Primary bandwidth use. Writes to the non-Primary site likewise have an increase in latency of 1x the RTT between sites (for a total of 2x the RTT) over writes to the Primary site.
- 如果这样的话，那除非另一站点完全故障了，否则怎么反转？那HYPERSWAP干嘛介绍这个？你的红皮书是哪来的？详见：From an initially created HyperSwap volume, the master vdisk will act as the primary. If there is 10 minutes of more I/O being submitted to the auxiliary vdisk’s site, the system will switch the direction of the relationship. From outside the system, the only visible effects of this are that the hosts on that site will have improved read and write performance, and the active-active relationship for that HyperSwap volume will have a primary attribute of “aux”. HyperSwap volumes in consistency groups all switch direction together. So the direction that a set of active-active relationships in a consistency group will replicate will depend on which of the two sites has the majority of host I/O across all HyperSwap volumes. Note that “the majority of I/O” is currently a comparison of number of sectors written to rather than a count of IOs. A 75% majority is required to switch to provide some hysteresis and prevent frequent alternating of direction.
- If there is 10 minutes of more I/O being submitted to the auxiliary 见“more I/O”，更多的IO，说明什么？说明更多的I/O被对端站点提供了，连续10分钟，大于75%，就会反转。 可能你理解错了，更多的I/O不是主机到SVC的，是SVC处理的，也就是SVC和存储处理的。
- 我觉得是你理解错了。being submitted to the auxiliary,应该是being submitted from the auxiliary site. 监控的是上层host SVC节点的访问流量。当跨站点访问流量更多时，发生切换以提高性能。试问，如你所说master被优先读写，那当什么情况下会出现aux访问超过75%呢？ 链路抖动？ 我的观点是SVC hyperswap的两站SVC间非global缓存，没有全局缓存的锁机制，两端各自的lun是不可能同时被读写的。另外，你对change volume的理解也是错的。change volume仅在双活关系断开又恢复时，重新同步时才会启用。正常状态下，是不参与双活数据传输的。
- 如果是两个Node分开部署，其实就相当于存储的双控拉开。此情况下两节点后端存储都可以同时读写。但是有单点失效的问题。 如果是两对Node分开部署，只能像您说的设成主从模式，从站点要想访问数据，必须跳转到主节点的那一对Node上。