diff --git a/src/dump.cpp b/src/dump.cpp
index 7c7f2cd3a2..2947b4c3fb 100644
--- a/src/dump.cpp
+++ b/src/dump.cpp
@@ -90,19 +90,25 @@ Dump::Dump(LAMMPS *lmp, int narg, char **arg) : Pointers(lmp)
   compressed = 0;
   binary = 0;
   multifile = 0;
+
   multiproc = 0;
+  nclusterprocs = nprocs;
+  filewriter = 0;
+  if (me == 0) filewriter = 1;
+  fileproc = 0;
+  multiname = NULL;
 
   char *ptr;
   if (ptr = strchr(filename,'%')) {
     multiproc = 1;
-    char *extend = new char[strlen(filename) + 16];
+    nclusterprocs = 1;
+    filewriter = 1;
+    fileproc = me;
+    MPI_Comm_split(world,me,0,&dumpcomm);
+    multiname = new char[strlen(filename) + 16];
     *ptr = '\0';
-    sprintf(extend,"%s%d%s",filename,me,ptr+1);
-    delete [] filename;
-    n = strlen(extend) + 1;
-    filename = new char[n];
-    strcpy(filename,extend);
-    delete [] extend;
+    sprintf(multiname,"%s%d%s",filename,me,ptr+1);
+    *ptr = '%';
   }
 
   if (strchr(filename,'*')) multifile = 1;
@@ -120,6 +126,7 @@ Dump::~Dump()
   delete [] id;
   delete [] style;
   delete [] filename;
+  delete [] multiname;
 
   delete [] format;
   delete [] format_default;
@@ -133,15 +140,15 @@ Dump::~Dump()
   memory->destroy(proclist);
   delete irregular;
 
+  if (multiproc) MPI_Comm_free(&dumpcomm);
+
   // XTC style sets fp to NULL since it closes file in its destructor
 
   if (multifile == 0 && fp != NULL) {
     if (compressed) {
-      if (multiproc) pclose(fp);
-      else if (me == 0) pclose(fp);
+      if (filewriter) pclose(fp);
     } else {
-      if (multiproc) fclose(fp);
-      else if (me == 0) fclose(fp);
+      if (filewriter) fclose(fp);
     }
   }
 }
@@ -167,6 +174,8 @@ void Dump::init()
   }
 
   if (sort_flag) {
+    if (multiproc > 1) 
+      error->all(FLERR,"Cannot dump sort when multiple procs write the dump file");
     if (sortcol == 0 && atom->tag_enable == 0)
       error->all(FLERR,"Cannot dump sort on atom IDs with no atom IDs defined");
     if (sortcol && sortcol > size_one)
@@ -266,27 +275,30 @@ void Dump::write()
     boxyz = domain->yz;
   }
 
-  // nme = # of dump lines this proc will contribute to dump
+  // nme = # of dump lines this proc contributes to dump
 
   nme = count();
-  bigint bnme = nme;
 
-  // ntotal = total # of dump lines
+  // ntotal = total # of dump lines in snapshot
   // nmax = max # of dump lines on any proc
 
+  bigint bnme = nme;
+  MPI_Allreduce(&bnme,&ntotal,1,MPI_LMP_BIGINT,MPI_SUM,world);
+
   int nmax;
-  if (multiproc) nmax = nme;
-  else {
-    MPI_Allreduce(&bnme,&ntotal,1,MPI_LMP_BIGINT,MPI_SUM,world);
-    MPI_Allreduce(&nme,&nmax,1,MPI_INT,MPI_MAX,world);
-  }
+  if (multiproc != nprocs) MPI_Allreduce(&nme,&nmax,1,MPI_INT,MPI_MAX,world);
+  else nmax = nme;
 
   // write timestep header
+  // for multiproc, nheader = # of lines in this file via Allreduce on dumpcomm
 
-  if (multiproc) write_header(bnme);
-  else write_header(ntotal);
+  bigint nheader = ntotal;
+  if (multiproc)
+    MPI_Allreduce(&bnme,&nheader,1,MPI_LMP_BIGINT,MPI_SUM,dumpcomm);
 
-  // insure proc 0 can receive everyone's info
+  if (filewriter) write_header(nheader);
+
+  // insure filewriter proc can receive everyone's info
   // limit nmax*size_one to int since used as arg in MPI_Rsend() below
   // pack my data into buf
   // if sorting on IDs also request ID list from pack()
@@ -309,46 +321,40 @@ void Dump::write()
   else pack(NULL);
   if (sort_flag) sort();
 
-  // multiproc = 1 = each proc writes own data to own file
-  // multiproc = 0 = all procs write to one file thru proc 0
-  //   proc 0 pings each proc, receives it's data, writes to file
-  //   all other procs wait for ping, send their data to proc 0
+  // filewriter = 1 = this proc writes to file
+  //   ping each proc in my cluster, receive its data, write data to file
+  // else wait for ping from fileproc, send my data to fileproc
 
-  if (multiproc) write_data(nme,buf);
-  else {
-    int tmp,nlines;
-    MPI_Status status;
-    MPI_Request request;
+  int tmp,nlines;
+  MPI_Status status;
+  MPI_Request request;
 
-    if (me == 0) {
-      for (int iproc = 0; iproc < nprocs; iproc++) {
-        if (iproc) {
-          MPI_Irecv(buf,maxbuf*size_one,MPI_DOUBLE,iproc,0,world,&request);
-          MPI_Send(&tmp,0,MPI_INT,iproc,0,world);
-          MPI_Wait(&request,&status);
-          MPI_Get_count(&status,MPI_DOUBLE,&nlines);
-          nlines /= size_one;
-        } else nlines = nme;
-
-        write_data(nlines,buf);
-      }
-      if (flush_flag) fflush(fp);
-
-    } else {
-      MPI_Recv(&tmp,0,MPI_INT,0,0,world,&status);
-      MPI_Rsend(buf,nme*size_one,MPI_DOUBLE,0,0,world);
+  if (filewriter) {
+    for (int iproc = 0; iproc < nclusterprocs; iproc++) {
+      if (iproc) {
+	MPI_Irecv(buf,maxbuf*size_one,MPI_DOUBLE,me+iproc,0,world,&request);
+	MPI_Send(&tmp,0,MPI_INT,me+iproc,0,world);
+	MPI_Wait(&request,&status);
+	MPI_Get_count(&status,MPI_DOUBLE,&nlines);
+	nlines /= size_one;
+      } else nlines = nme;
+      
+      write_data(nlines,buf);
     }
+    if (flush_flag) fflush(fp);
+    
+  } else {
+    MPI_Recv(&tmp,0,MPI_INT,fileproc,0,world,&status);
+    MPI_Rsend(buf,nme*size_one,MPI_DOUBLE,fileproc,0,world);
   }
 
-  // if file per timestep, close file
+  // if file per timestep, close file if I am filewriter
 
   if (multifile) {
     if (compressed) {
-      if (multiproc) pclose(fp);
-      else if (me == 0) pclose(fp);
+      if (filewriter) pclose(fp);
     } else {
-      if (multiproc) fclose(fp);
-      else if (me == 0) fclose(fp);
+      if (filewriter) fclose(fp);
     }
   }
 }
@@ -368,27 +374,29 @@ void Dump::openfile()
 
   // if one file per timestep, replace '*' with current timestep
 
-  char *filecurrent;
-  if (multifile == 0) filecurrent = filename;
-  else {
-    filecurrent = new char[strlen(filename) + 16];
-    char *ptr = strchr(filename,'*');
+  char *filecurrent = filename;
+  if (multiproc) filecurrent = multiname;
+
+  if (multifile) {
+    char *filestar = filecurrent;
+    filecurrent = new char[strlen(filestar) + 16];
+    char *ptr = strchr(filestar,'*');
     *ptr = '\0';
     if (padflag == 0)
       sprintf(filecurrent,"%s" BIGINT_FORMAT "%s",
-              filename,update->ntimestep,ptr+1);
+              filestar,update->ntimestep,ptr+1);
     else {
       char bif[8],pad[16];
       strcpy(bif,BIGINT_FORMAT);
       sprintf(pad,"%%s%%0%d%s%%s",padflag,&bif[1]);
-      sprintf(filecurrent,pad,filename,update->ntimestep,ptr+1);
+      sprintf(filecurrent,pad,filestar,update->ntimestep,ptr+1);
     }
     *ptr = '*';
   }
 
-  // open one file on proc 0 or file on every proc
+  // each proc with filewriter = 1 opens a file
 
-  if (me == 0 || multiproc) {
+  if (filewriter) {
     if (compressed) {
 #ifdef LAMMPS_GZIP
       char gzip[128];
@@ -552,8 +560,7 @@ void Dump::sort()
   // this insures proc 0 can receive everyone's info
 
   int nmax;
-  if (multiproc) nmax = nme;
-  else MPI_Allreduce(&nme,&nmax,1,MPI_INT,MPI_MAX,world);
+  MPI_Allreduce(&nme,&nmax,1,MPI_INT,MPI_MAX,world);
 
   if (nmax > maxbuf) {
     maxbuf = nmax;
@@ -669,6 +676,36 @@ void Dump::modify_params(int narg, char **arg)
       else if (strcmp(arg[iarg+1],"no") == 0) first_flag = 0;
       else error->all(FLERR,"Illegal dump_modify command");
       iarg += 2;
+
+    } else if (strcmp(arg[iarg],"fileper") == 0) {
+      if (iarg+2 > narg) error->all(FLERR,"Illegal dump_modify command");
+      if (!multiproc)
+	error->all(FLERR,"Cannot use dump_modify fileper "
+                   "without % in dump file name");
+      int nper = atoi(arg[iarg+1]);
+      if (nper <= 0) error->all(FLERR,"Illegal dump_modify command");
+      
+      multiproc = nprocs/nper;
+      if (nprocs % nper) multiproc++;
+      fileproc = me/nper * nper;
+      int fileprocnext = MIN(fileproc+nper,nprocs);
+      nclusterprocs = fileprocnext - fileproc;
+      if (me == fileproc) filewriter = 1;
+      else filewriter = 0;
+      int icluster = fileproc/nper;
+
+      MPI_Comm_free(&dumpcomm);
+      MPI_Comm_split(world,icluster,0,&dumpcomm);
+
+      delete [] multiname;
+      multiname = new char[strlen(filename) + 16];
+      char *ptr = strchr(filename,'%');
+      *ptr = '\0';
+      sprintf(multiname,"%s%d%s",filename,icluster,ptr+1);
+      *ptr = '%';
+      iarg += 2;
+
+
     } else if (strcmp(arg[iarg],"flush") == 0) {
       if (iarg+2 > narg) error->all(FLERR,"Illegal dump_modify command");
       if (strcmp(arg[iarg+1],"yes") == 0) flush_flag = 1;
@@ -685,6 +722,40 @@ void Dump::modify_params(int narg, char **arg)
         strcpy(format_user,arg[iarg+1]);
       }
       iarg += 2;
+
+    } else if (strcmp(arg[iarg],"nfile") == 0) {
+      if (iarg+2 > narg) error->all(FLERR,"Illegal dump_modify command");
+      if (!multiproc)
+	error->all(FLERR,"Cannot use dump_modify nfile "
+                   "without % in dump file name");
+      int nfile = atoi(arg[iarg+1]);
+      if (nfile <= 0) error->all(FLERR,"Illegal dump_modify command");
+      nfile = MIN(nfile,nprocs);
+
+      multiproc = nfile;
+      int icluster = static_cast<int> ((bigint) me * nfile/nprocs);
+      fileproc = static_cast<int> ((bigint) icluster * nprocs/nfile);
+      int fcluster = static_cast<int> ((bigint) fileproc * nfile/nprocs);
+      if (fcluster < icluster) fileproc++;
+      int fileprocnext = 
+        static_cast<int> ((bigint) (icluster+1) * nprocs/nfile);
+      fcluster = static_cast<int> ((bigint) fileprocnext * nfile/nprocs);
+      if (fcluster < icluster+1) fileprocnext++;
+      nclusterprocs = fileprocnext - fileproc;
+      if (me == fileproc) filewriter = 1;
+      else filewriter = 0;
+
+      MPI_Comm_free(&dumpcomm);
+      MPI_Comm_split(world,icluster,0,&dumpcomm);
+
+      delete [] multiname;
+      multiname = new char[strlen(filename) + 16];
+      char *ptr = strchr(filename,'%');
+      *ptr = '\0';
+      sprintf(multiname,"%s%d%s",filename,icluster,ptr+1);
+      *ptr = '%';
+      iarg += 2;
+
     } else if (strcmp(arg[iarg],"pad") == 0) {
       if (iarg+2 > narg) error->all(FLERR,"Illegal dump_modify command");
       padflag = atoi(arg[iarg+1]);
diff --git a/src/dump.h b/src/dump.h
index b54705aa93..06b331d113 100644
--- a/src/dump.h
+++ b/src/dump.h
@@ -14,6 +14,7 @@
 #ifndef LMP_DUMP_H
 #define LMP_DUMP_H
 
+#include "mpi.h"
 #include "stdio.h"
 #include "pointers.h"
 
@@ -55,7 +56,14 @@ class Dump : protected Pointers {
   int compressed;            // 1 if dump file is written compressed, 0 no
   int binary;                // 1 if dump file is written binary, 0 no
   int multifile;             // 0 = one big file, 1 = one file per timestep
-  int multiproc;             // 0 = proc 0 writes for all, 1 = one file/proc
+
+  int multiproc;             // 0 = proc 0 writes for all, 
+                             // else # of procs writing files
+  int nclusterprocs;         // # of procs in my cluster that write to one file
+  int filewriter;            // 1 if this proc writes a file, else 0
+  int fileproc;              // ID of proc in my cluster who writes to file
+  char *multiname;           // dump filename with % converted to cluster ID
+  MPI_Comm dumpcomm;
 
   int header_flag;           // 0 = item, 2 = xyz
   int flush_flag;            // 0 if no flush, 1 if flush every dump
@@ -80,7 +88,7 @@ class Dump : protected Pointers {
   double boxzlo,boxzhi;
   double boxxy,boxxz,boxyz;
 
-  bigint ntotal;             // # of per-atom lines in snapshot
+  bigint ntotal;             // total # of per-atom lines in snapshot
   int reorderflag;           // 1 if OK to reorder instead of sort
   int ntotal_reorder;        // # of atoms that must be in snapshot
   int nme_reorder;           // # of atoms I must own in snapshot