From: arne@vajhoej.dk   
      
   On 11/13/2025 5:36 AM, David Wade wrote:   
   > On 13/11/2025 02:45, Lawrence D’Oliveiro wrote:   
   >> On Wed, 12 Nov 2025 21:43:40 +0000, David Wade wrote:   
   >>> Why won't it scale to 460,000 nodes?   
   >>   
   >> Because a cluster of on the order of tens of machines (like your SysPlex   
   >> and VMScluster) can depend on algorithms with polynomial complexity, that   
   >> would no longer be practicable when you have hundreds of thousands of   
   >> nodes.   
   >>   
   >>> Why would you need that many nodes, well unless you are google?   
   >>   
   >> All the hyperscalers are running clusters of that sort of size.   
   >   
   > Are they really tightly couple clusters, or load balanced front ends...   
      
   Note that the hundreds of thousands of nodes are all clusters   
   not one cluster.   
      
   But none of the clusters will be OS clusters like z/OS   
   SysPlex or VMS cluster. Some form of application cluster.   
      
   Companies of that size are all unique, but let us invent   
   a hypothetical company SuperBigBiz.   
      
   Transaction side:   
      
   customers   
        |   
        v   
   load balancer with sticky sessions   
        |   
        v   
   2000 node.js instances stateless + 2000 readonly copies of static content   
        |   
        v   
   load balancer   
        |   
        v   
   5000 Java SpringBoot micro-service instances stateless   
   smart client libraries   
        |   
        v   
   100 sharded Redis instances + 1000 sharded PostgreSQL instances   
      
   Analytical side:   
      
   Python loaders running on all nodes   
   smart client library   
        |   
        v   
   10 sharded Kafka instances   
        ^   
        |   
   100 Python loader instances   
   smart client library   
        |   
        v   
   5000 sharded Cassandra instances   
        ^   
        |   
   smart client library   
   1000 Spark instances splitting work   
        ^   
        |   
   smart client library   
   1 PC querying data   
      
   That gives 14210 instances (VM or container).   
      
   Much smaller than Google, FaceBook etc., but still bigger than   
   what you would do with z/OS or VMS.   
      
   Also most large companies have multiple different transactional flows.   
      
   Arne   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)