[GitHub] [incubator-mxnet] anandj91 commented on a change in pull request #15124: [MXNET-1294] Priority-based parameter propagation for improved data parallel training throughput

[GitBox <gi...@apache.org>]

anandj91 commented on a change in pull request #15124: [MXNET-1294] Priority-based parameter propagation for improved data parallel training throughput
URL: https://github.com/apache/incubator-mxnet/pull/15124#discussion_r291472650
 
 

 ##########
 File path: src/kvstore/kvstore_dist.h
 ##########
 @@ -544,31 +592,25 @@ class KVStoreDist : public KVStoreLocal {
       const int num_servers = krs.size();
       CHECK_GT(num_servers, 0);
 
-      // a simple heuristic for load balance
-      if (num_arr_elems < bigarray_bound_) {
-        // send it to a single random picked server
-        int server = (key * 9973) % num_servers;
-        ps::Key ps_key = krs[server].begin() + key;
-        CHECK_LT(ps_key, krs[server].end());
+      /**
+       * Round-Robin key assignment
+       */
+      int64_t params = pskv_size;
+      int64_t slice_bound = bigarray_bound_ * num_bytes;
+      static ps::Key server = 0;
 
 Review comment:
   the server variable is static. so it is incremented on every invocation of EncodeDefaultKey function. this does not cause overloading server 0.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services