Hello, It's a bit old but at least for me, still a great guide: https://tobert.github.io/pages/als-cassandra-21-tuning-guide.html
My 2 cents: We deal with electronic invoices and our load is about 10,000 transactions/s during the peak housr. We are not located in USA, so AWS would be a bit expensive for our project. After a lot of research, tests and simulations (technical and financial) I decide to purchase four Supermicros MicroClouds 5038ML-H12TRF with Xeon E3 series, 32GB RAM and 4 x 2.5 "1TB spinning disks each node. The chassis are divided in 2 DCs with 2 x 1Gbps links redundant for DC/DC interconnection and each blade is divided as follow: 3 x Cassandra 2.0 nodes for production, 3 x Cassandra 3.x (tests for migration), 3 x Spark / Hadoop for analytics and 3 x application servers. So, we have a 12 node Cassandra 2.0 that has been working very fine for the last 3 years with a very low latency and overhead. Some bumps here and there but with properly management and monitoring we can deal with almost everything. Despite we use 2.5 "disks, we always check the BlackBlaze's hard drive reliability reports before any disk purchasing: https://www.backblaze.com/blog/hard-drive-failure-rates-q1-2017/ On Cas 2.0 we started using separeted disks for the data_file_directories. On Cas 3.x, following Al Tobey's guide, we're using MD Raid0 in a XFS filesystem and the performance are far better than on Cas 2.0. I hope it helps. Jero On Fri, Sep 29, 2017 at 3:19 AM, Peng Xiao <2535...@qq.com> wrote: > Hi there, > we are struggling on hardware selection,we all know that ssd is good,and > Datastax suggests us to use ssd,as Cassandra is a CPU bound db,we are > considering to use sata disk,we noticed that the normal IO throughput is > 7MB/s. > > Could anyone give some advice? > > Thanks, > Peng Xiao > >