Index: client.scm
==================================================================
--- client.scm
+++ client.scm
@@ -92,19 +92,23 @@
 		      (tasks:server-force-clean-run-record (db:delay-if-busy tdbdat)
 							   run-id 
 							   (tasks:hostinfo-get-interface server-dat)
 							   (tasks:hostinfo-get-port      server-dat)
 							   " client:setup (server-dat = #t)")
+		      (if (> remaining-tries 8)
+			  (thread-sleep! (+ 1 (random 5))) ;; spread out the starts a little
+			  (thread-sleep! (+ 15 (random 20)))) ;; it isn't going well. give it plenty of time
 		      (server:try-running run-id)
-		      (thread-sleep! 5) ;; give server a little time to start up
-		      (client:setup run-id remaining-tries: (- remaining-tries 1)))))
+		      (thread-sleep! 5)   ;; give server a little time to start up
+		      (client:setup run-id remaining-tries: (- remaining-tries 1))
+		      )))
 	      (begin    ;; no server registered
 		(let ((num-available (tasks:num-in-available-state (db:dbdat-get-db tdbdat) run-id)))
 		  (debug:print-info 0 "client:setup, no server registered, remaining-tries=" remaining-tries " num-available=" num-available)
 		  (if (< num-available 2)
 		      (server:try-running run-id))
-		  (thread-sleep! 5) ;; give server a little time to start up
+		  (thread-sleep! (+ 5 (random (- 20 remaining-tries))))  ;; give server a little time to start up, randomize a little to avoid start storms.
 		  (client:setup run-id remaining-tries: (- remaining-tries 1)))))))))
 
 ;; 	(let ((host-info (hash-table-ref/default *runremote* run-id #f)))
 ;; 	  (if host-info ;; this is a bit circular. the host-info *is* the start-res FIXME
 ;; 	      (let* ((iface     (http-transport:server-dat-get-iface host-info))

Index: http-transport.scm
==================================================================
--- http-transport.scm
+++ http-transport.scm
@@ -287,11 +287,13 @@
 					     (set! success #f)
 					     (debug:print 0 "WARNING: failure in with-input-from-request to " fullurl ".")
 					     (debug:print 0 " message: " ((condition-property-accessor 'exn 'message) exn))
 					     (hash-table-delete! *runremote* run-id)
 					     ;; Killing associated server to allow clean retry.")
-					     ;; (tasks:kill-server-run-id run-id)  ;; better to kill the server in the logic that called this routine.
+					     (tasks:kill-server-run-id run-id)  ;; better to kill the server in the logic that called this routine?
+					     (signal (make-composite-condition
+						      (make-property-condition 'commfail 'message "failed to connect to server")))
 					     #f)
 					   (with-input-from-request ;; was dat
 					    fullurl 
 					    (list (cons 'key "thekey")
 						  (cons 'cmd cmd)

Index: rmt.scm
==================================================================
--- rmt.scm
+++ rmt.scm
@@ -87,11 +87,13 @@
 	 (connection-info (rmt:get-connection-info run-id)))
     ;; the nmsg method does the encoding under the hood (the http method should be changed to do this also)
     (if connection-info
 	;; use the server if have connection info
 	(let* ((dat     (case *transport-type*
-			  ((http)(http-transport:client-api-send-receive run-id connection-info cmd params))
+			  ((http)(condition-case
+				  (http-transport:client-api-send-receive run-id connection-info cmd params)
+				  ((commfail)(vector #f "communications fail"))))
 			  ((nmsg)(condition-case
 				  (nmsg-transport:client-api-send-receive run-id connection-info cmd params)
 				  ((timeout)(vector #f "timeout talking to server"))))
 			  (else  (exit))))
 	       (success (if (and dat (vector? dat)) (vector-ref dat 0) #f))