Hey Vishnu, I listened to your webinar. It was nice to hear it. One thing I was wondering was that you mentioned that for benchmarking we should not use the production servers, but rather have a replica of them. My question is that, with replica we can do only limited types of benchmarking viz. capacity testing. But testing of cache hits can be done only on production servers where we have real queries from user. What say on that ?
Hi Vishal, Suggestions are more about what to avoid, doesn't mean mustn't be done. For benchmarkIf need not to be on exact real users traffic, using different simulations it should be done. Some alternatives: 1. For the cases where virtual traffic can be generated with similar to prod data on testing environment, it should be run on testing server only. 2. If you want to run on production only, and you can take that particular server out of rotation if possible for the benchmarking 3. On live production during non-peak hours, or may be smaller set (bucket) of users.
Off course there will be cases where you are not left but doing the benchmark on production. Those cases should be minimal to avoid any impact on real user base.
2 comments:
Hey Vishnu, I listened to your webinar. It was nice to hear it. One thing I was wondering was that you mentioned that for benchmarking we should not use the production servers, but rather have a replica of them. My question is that, with replica we can do only limited types of benchmarking viz. capacity testing. But testing of cache hits can be done only on production servers where we have real queries from user. What say on that ?
Hi Vishal,
Suggestions are more about what to avoid, doesn't mean mustn't be done.
For benchmarkIf need not to be on exact real users traffic, using different simulations it should be done. Some alternatives:
1. For the cases where virtual traffic can be generated with similar to prod data on testing environment, it should be run on testing server only.
2. If you want to run on production only, and you can take that particular server out of rotation if possible for the benchmarking
3. On live production during non-peak hours, or may be smaller set (bucket) of users.
Off course there will be cases where you are not left but doing the benchmark on production. Those cases should be minimal to avoid any impact on real user base.
Post a Comment