[PVFS2-CVS] commit by neill in pvfs2/src/client/sysint: client-state-machine.h sys-io.sm

[CVS commit program]

Update of /projects/cvsroot/pvfs2/src/client/sysint
In directory parlweb:/tmp/cvs-serv23203/src/client/sysint

Modified Files:
	client-state-machine.h sys-io.sm 
Log Message:
- introduce a scaling flow timeout for large I/O operations
- modify the write_ack timeout to be a bit longer than the computed
  flow timeout


Index: client-state-machine.h
===================================================================
RCS file: /projects/cvsroot/pvfs2/src/client/sysint/client-state-machine.h,v
diff -p -u -r1.120 -r1.121
--- client-state-machine.h	13 Jul 2004 02:14:18 -0000	1.120
+++ client-state-machine.h	13 Jul 2004 15:29:05 -0000	1.121
@@ -136,6 +136,7 @@ typedef struct
     /* a reference to the msgpair we're using for communication */
     PINT_sm_msgpair_state *msg;
 
+    int flow_timeout;
     job_id_t flow_job_id;
     job_status_s flow_status;
     flow_descriptor flow_desc;

Index: sys-io.sm
===================================================================
RCS file: /projects/cvsroot/pvfs2/src/client/sysint/sys-io.sm,v
diff -p -u -r1.84 -r1.85
--- sys-io.sm	13 Jul 2004 02:14:18 -0000	1.84
+++ sys-io.sm	13 Jul 2004 15:29:05 -0000	1.85
@@ -57,6 +57,8 @@ static inline int process_context_recv_a
 static inline int check_context_status(
     PINT_client_io_ctx *cur_ctx, int io_type,
     PVFS_size *total_size);
+static inline int compute_scaling_flow_timeout(
+    PVFS_size data_size, int datafile_count);
 static int io_find_target_datafiles(
     PVFS_Request mem_req, PVFS_Request file_req,
     PVFS_offset file_req_offset, PINT_dist *dist_p,
@@ -1205,20 +1207,25 @@ static inline int build_context_flow(
     cur_ctx->flow_desc.file_data.fsize = resp->u.io.bstream_size;
     cur_ctx->flow_desc.file_data.dist = attr->u.meta.dist;
     cur_ctx->flow_desc.file_data.server_nr = cur_ctx->server_nr;
-    cur_ctx->flow_desc.file_data.server_ct = attr->u.meta.dfile_count;
+    cur_ctx->flow_desc.file_data.server_ct = sm_p->u.io.datafile_count;
 
     cur_ctx->flow_desc.file_req = sm_p->u.io.file_req;
     cur_ctx->flow_desc.file_req_offset = sm_p->u.io.file_req_offset;
 
     cur_ctx->flow_desc.mem_req = sm_p->u.io.mem_req;
 
+    cur_ctx->flow_timeout = compute_scaling_flow_timeout(
+        cur_ctx->flow_desc.file_data.fsize,
+        cur_ctx->flow_desc.file_data.server_ct);
+
     gossip_debug(
         GOSSIP_IO_DEBUG, "    bstream_size = %Ld, datafile_nr = "
-        "%d, datafile_ct = %d, file_req_off = %Ld\n",
+        "%d, datafile_ct = %d\n\tfile_req_off = %Ld (timeout=%d secs)\n",
         Ld(cur_ctx->flow_desc.file_data.fsize),
         cur_ctx->flow_desc.file_data.server_nr,
         cur_ctx->flow_desc.file_data.server_ct,
-        Ld(cur_ctx->flow_desc.file_req_offset));
+        Ld(cur_ctx->flow_desc.file_req_offset),
+        cur_ctx->flow_timeout);
 
     cur_ctx->flow_desc.tag = cur_ctx->session_tag;
     cur_ctx->flow_desc.type = sm_p->u.io.flowproto_type;
@@ -1328,6 +1335,9 @@ static inline int process_context_recv_a
         /*
           we're pre-posting the final write ack here, even though it's
           ahead of the flow phase; reads are at the flow phase.
+
+          the timeout used here is slightly longer than the flow
+          (since this should happen sometime after the flow completes)
         */
         status_user_tag = ((4 * cur_ctx->index) + IO_SM_PHASE_FINAL_ACK);
 
@@ -1336,7 +1346,8 @@ static inline int process_context_recv_a
             cur_ctx->write_ack.max_resp_sz, cur_ctx->session_tag,
             BMI_PRE_ALLOC, sm_p, status_user_tag,
             &cur_ctx->write_ack.recv_status, &cur_ctx->write_ack.recv_id,
-            pint_client_sm_context, PVFS2_CLIENT_JOB_TIMEOUT);
+            pint_client_sm_context,
+            (cur_ctx->flow_timeout + PVFS2_CLIENT_JOB_TIMEOUT));
 
         if (ret < 0)
         {
@@ -1355,7 +1366,7 @@ static inline int process_context_recv_a
     ret = job_flow(
         &cur_ctx->flow_desc, sm_p, status_user_tag,
         &cur_ctx->flow_status, &cur_ctx->flow_job_id,
-        pint_client_sm_context, PVFS2_CLIENT_JOB_TIMEOUT);
+        pint_client_sm_context, cur_ctx->flow_timeout);
 
     if (ret < 0)
     {
@@ -1478,6 +1489,16 @@ static inline int check_context_status(
         }
     }
     return ret;
+}
+
+static inline int compute_scaling_flow_timeout(
+    PVFS_size data_size, int datafile_count)
+{
+    static int one_hundred_megs = (1024 * 1024 * 100);
+    int scale = ((data_size / one_hundred_megs) * datafile_count);
+
+    return (scale ? (scale * PVFS2_CLIENT_JOB_TIMEOUT) :
+            PVFS2_CLIENT_JOB_TIMEOUT);
 }
 
 /*