You can pretty much run as many frameworks as you'd like. One trick you can use to drive up CPU use (if you want) is to overcommit CPUs. For example, if you have a machine with 16 CPUs, set the number of CPUs to 32 on the Mesos agent, and you'll be overcommitted by 100% (i.e., 1 CPU share will practically be equivalent to 50% of 1 CPU).
Furthermore, you will want to ensure that allocated resources match *used* resources, which may help drive up utilization if that's your goal (aka 'right sizing'). On Thu, Jul 21, 2016 at 9:35 PM, Sam <[email protected]> wrote: > Hi Guys, > We met one issue in maximizing the CPU and Mem resources utilization in > our servers. For example, We got 100 CPUs and 100G Mem , and we installed > Apache mesos and Kafka \ Hbase frameworks. In our production Env, we got > heavy workload from client in Hbase/Kafka, however our resource utilization > percentage is always under 50%. it's so surprised that Mesos cannot be able > to offer the full resources to frameworks. > So we are thinking of deploy same multi-Frameworks to fix this issue, for > Instance, 5 Kafka frameworks to allocate at least 20% of resources, that's > 20 CPUs and 20MEMs each. Which me mans 5 Kafkas will consume 100CPUs and > 100MEMs logically. Let's say using role, quota or reservation of this is > the case. > > Our questions are : > 1, Is this logic correct? Or ridiculous? > 2, if this is positive, how to support and customize frameworks or others > to support this? Which means I can order 5 times in same cluster. > > Looking forward to have any suggestions that you got. > Appreciated > > Regards, > Sam > > Sent from my iPhone
