卡珊德拉集群的插入性能差，插入稳定性差

I have to store around 250 numerical values per second, per client, which is around 900k numbers per hour. It probably will not be a full-day recording (probably between 5-10 hours a day), but I will partition my data based on the client id and the day the reading is made. The maximum row length comes at about 22-23M which is still manageable. Neverteless, my scheme looks like this:I have to store around 250 numerical values per