3-node cluster recovery in case of one node failure
Pages: 1 2
admin
2,930 Posts
March 27, 2018, 9:01 amQuote from admin on March 27, 2018, 9:01 amThe recovery speed is determined by 3 things:
- Your cluster performance
- Your client io load
- Settings to give priority of client io vs recovery io
You should always plan to have enough cluster performance to handle your "real" client workload + recovery operation at the same time. This is specially important in small clusters, where a node down could mean 25% of your cluster needs to be recovered. You should perform such tests before production and get an idea on how to scale your nodes if your client io increases. Also it is important to use "real" expected load...many of us when running benchmarks, we stress the cluster to its knees to get the max performance, obviously a maxed out cluster at peek load will not be able to handle additional recovery traffic. Changing client/recovery priority is not the best solution and is case by case dependent. it is best to have good performance.
The recovery speed is determined by 3 things:
- Your cluster performance
- Your client io load
- Settings to give priority of client io vs recovery io
You should always plan to have enough cluster performance to handle your "real" client workload + recovery operation at the same time. This is specially important in small clusters, where a node down could mean 25% of your cluster needs to be recovered. You should perform such tests before production and get an idea on how to scale your nodes if your client io increases. Also it is important to use "real" expected load...many of us when running benchmarks, we stress the cluster to its knees to get the max performance, obviously a maxed out cluster at peek load will not be able to handle additional recovery traffic. Changing client/recovery priority is not the best solution and is case by case dependent. it is best to have good performance.
Last edited on March 27, 2018, 9:01 am by admin · #11
Pages: 1 2
3-node cluster recovery in case of one node failure
admin
2,930 Posts
Quote from admin on March 27, 2018, 9:01 amThe recovery speed is determined by 3 things:
- Your cluster performance
- Your client io load
- Settings to give priority of client io vs recovery io
You should always plan to have enough cluster performance to handle your "real" client workload + recovery operation at the same time. This is specially important in small clusters, where a node down could mean 25% of your cluster needs to be recovered. You should perform such tests before production and get an idea on how to scale your nodes if your client io increases. Also it is important to use "real" expected load...many of us when running benchmarks, we stress the cluster to its knees to get the max performance, obviously a maxed out cluster at peek load will not be able to handle additional recovery traffic. Changing client/recovery priority is not the best solution and is case by case dependent. it is best to have good performance.
The recovery speed is determined by 3 things:
- Your cluster performance
- Your client io load
- Settings to give priority of client io vs recovery io
You should always plan to have enough cluster performance to handle your "real" client workload + recovery operation at the same time. This is specially important in small clusters, where a node down could mean 25% of your cluster needs to be recovered. You should perform such tests before production and get an idea on how to scale your nodes if your client io increases. Also it is important to use "real" expected load...many of us when running benchmarks, we stress the cluster to its knees to get the max performance, obviously a maxed out cluster at peek load will not be able to handle additional recovery traffic. Changing client/recovery priority is not the best solution and is case by case dependent. it is best to have good performance.