I think newer high core servers are a perfect fit for vertical scaling of manticore search.

In the past when servers had few cores, I considered elastic search distributed system very advantageous because using low core servers with sphinx quickly saturated a single server, with the trade-off being that elastic search is much more complex to setup.

However, due to new servers with physical 128 cores already easily available, manticore search will probably be adequate for many workloads. We found that tuning max_threads_per_query settings in high core servers easily allows a tuning response time performance with the trade-off of a better response time needing increased server cpu consumption.

Due to the newer high core count servers that are going to be released, I think this is a very good fit for vertical scaling manticore search easily without complex server engineering and clustering setup.
I found it is possible to scale quite a lot vertically before needing more servers when using high end servers.
Often in my experience, this will not only be cheaper but faster than than managing a cluster of servers.

We have been using a 128 core (256 thread) server for sphinx search 2.2 and recently migrated to manticore search with very good results.

Also we found that in a busy server environment manticore search 6 slightly decreased server load compared to sphinx 2.1 with almost the exact settings with a slightly improved response time.

Thank you, manticore search team.

1 Like