What's important for me was to do local caching on the user session, and the restore can be possible without reading DB and reconstruction. If the users will be redistributed, a separate caching server like memcache can be used.
多谢指点,豁然开朗!我还是没有系统设计经验,思维局限在单机。 有备份机memcache的话,可以对每个手机终端(IMEI)缓存ssthresh和cwnd, 理论上重启、TCP握手后,貌似server可以恢复到断之前的congestion windows size。 但client端TCP貌似还是要slow start?除非hack TCP协议。
【在 g*****g 的大作中提到】 : What's important for me was to do local caching on the user session, and : the restore can be possible without reading DB and reconstruction. If the : users will be : redistributed, a separate caching server like memcache can be used.
I think recovery on application layer is more than enough. If immediate recovery is critical, a failover strategy should be used instead. e.g. The next heartbeat/request can be redirected by load balancer to a different node and the user session can be recovered from there, instead of waiting on the failed node.
【在 g*****g 的大作中提到】 : I think recovery on application layer is more than enough. If immediate : recovery is critical, a failover strategy should be used instead. e.g. The : next heartbeat/request can be redirected by load balancer to a different : node and the user : session can be recovered from there, instead of waiting on the failed node.