We need benchmarks or data to support the claim. A single OS and JVM does not scale linearly for higher loads. If you have seperate OS and Java, the load is distributed across multiple instances (with each instance only requiered to support a smaller load and hence would scale nicely)
I had found this for running multiple apache servers on multiple VMs as compared to a single instance (not Solr). But i am pretty sure it would be same for Solr too On 12 Mar 2018 12:42, "Shawn Heisey" <apa...@elyograg.org> wrote: > On 3/11/2018 7:39 PM, Deepak Goel wrote: > >> I doubt this. It would be great if someone can subtantiate this with hard >> facts >> > > This seems to be in response to my claim that virtualization always has > overhead. I don't see how this statement can be at all controversial. > > Virtualization isn't free, even if the hardware and software in use are > extremely efficient at it. Translating what a virtual machine does into a > corresponding action on the real hardware is going to take time and > resources beyond whatever the action itself is. > > Plus there's the application-level overhead. You have the overhead of > multiple operating systems, multiple copies of Java running, multiple > servlet containers (probably Jetty), and multiple copies of Solr. And each > of them is running inside a limited subset of the hardware installed in the > physical server. > > Let's say you start with VMs on a server, and benchmark Solr's > performance. Then you completely erase the server, install one operating > system, install Solr onto the OS, and then install all of the indexes that > were running on the VMs into that one Solr instance. Assuming that things > are set up correctly and that you give that Solr instance the correct > amount of heap memory, it's almost guaranteed to be faster than the VMs. I > can't tell you whether the improvement will be half a percent or 50 > percent, only that it will be faster. > > Thanks, > Shawn > >
