Times for nt=1024 nx=256 on the NCUBE with standard program, single user,
Crystalline Operating System emulation.  7/5/87  -  Stewart A. Levin

ntau	start	end	diff
1	8:56:30	9:02:50	6:20
2	9:06:35	9:14:35	8:00
3	8:35:43	8:45:20	9:37
4	9:18:45	9:30:02	11:17
6	9:33:15	9:48:18	15:03


Times for nt=1024 nx=256 with program modified for m circuits through cube.
Packet size fixed at 8 (doubleword).

ntau	m	start	end	diff	notes
1	1	1:00:45	1:07:30	6:45	rotating mshift, host outside loop
1	1	0:33:25	0:40:10	6:45	fixed mshift, host outside loop
1	1	1:34:50	1:39:55	5:05	fixed mshift, host in loop
2	1	1:44:07	1:51:00	6:53	""
1	2	2:05:07	2:13:08	8:01	""
3	1	2:18:10	2:27:00	8:50	""
1	3	2:34:00	2:46:15	12:15	""
1	3	9:35:55	9:49:15	13:20	copy packets to avoid potential r/w. confl.
1	3	1:53:06	2:06:24	13:18	rotating mshift
1	3	0:09:21	0:22:36	13:15	rotating mshift, parity flip/flop
1	3	1:04:00	1:16:16	12:16	"" recode for no packet copy


Times for nt=1024 nx=256 with varying packet sizes.  Synchronization all write
then read.  (Jammed up other way once when I reversed the order of the circuit!)

pktlen	ntau	m	rotate	greedy	mildly
8	1	3	12:17	14:04
8	3	1	8:18	8:27
16	1	3	8:34	9:27
16	3	1	7:12	6:48
32	1	3	6:41	7:07
32	3	1	6:00	5:45
64	1	3	5:44	5:57
64	3	1	5:28	5:29
256	1	3	5:07	5:08
256	3	1	5:13	5:13
1024	1	3	4:59	4:59
1024	3	1	lockup	lockup

Times for nt=1024 nx=256 packet=256 for varying cube dimension, ntau, m.
These times are slower than above because input data is not mostly zero,
which I find is special cased by NCUBE.

dim	ntau	m	time
6	3	1	8:44
6	1	3	8:32
6	2	1	5:53
6	1	2	5:53
6	1	1	3:00
4	3	1	7:01
4	2	1	4:44
4	1	2	4:47
4	1	1	2:25
2	3	1	6:55
2	2	1	4:38
2	1	1	2:22
0	3	1	6:53
0	2	1	4:38
0	1	1	2:22