Here are the results of test I ran to transfer a 32MB file. In the tests, the ssh server always ran on the W2K3sp1 machine. The 32MB file consisted of 1024 * 1024 * 32 spaces. As such I would have though compression should have given some phenomenal results. Surprisingly it didn't... except in the case of the copSSH-HPN client against copSSH-HPN server.
- W2K3sp1 w/ dual 1263MHz CPUs and 512MB RAM
- XPsp2 w/ 433MHz CPU and 128MB RAM
Copying the 32 MB file from the command line using copy and UNC paths (i.e. w/o ssh) gave the following results:
WinXPsp2 --> W2K3sp1 (put):
10.576 secs (3098.312 KB/s)
9.644 secs (3397.924 KB/s)
WinXPsp2
To Garrett:
I tried to update topic title. Unfortunately, BBS software removed almost all content in your posting. Terribly sorry for that. Is it possible to repost your findings ?
Rgrds Tev
A comparative test:client - XP SP1, server - XP SP1, connected to a dedicated 100 Mbps switched network, a 855 MB ghost image file via command 'scp -i identity file user@server:'
#2: 242 secs
#3: 242 secs
#2: 276 secs
TCP/IP window sizes as recommended
#2: 270 secs
It seems to me there is a bottleneck somewhere.
SSL performance test results copying 32MB file consisting of 1024 * 1024 * 32 spaces
9.644 secs (3397.924 KB/s)
4.949 secs (6633.956 KB/s)
81.150 secs (403.798 KB/s)
60.201 secs (544.305 KB/s)
30.124 secs (1087.771 KB/s)
21.479 secs (1525.570 KB/s)
20.955 secs (1563.761 KB/s)
28.639 secs (1144.178 KB/s)
20.495 secs (1598.834 KB/s)
78.404 secs (417.937 KB/s)
89.832 secs (364.768 KB/s)
71.352 secs (459.243 KB/s)
54.254 secs (603.973 KB/s)
12.570 secs (2606.755 KB/s)
71.639 secs (457.403 KB/s)
12.370 secs (2648.835 KB/s)
293.477 secs (111.654 KB/s)
291.943 secs (112.241 KB/s)
421.606 secs (77.722 KB/s)
28.284 secs (1158.549 KB/s)
27.301 secs (1200.266 KB/s)
453.295 secs (72.288 KB/s)
30.879 secs (1061.185 KB/s)
98.584 secs (332.387 KB/s)
100.626 secs (325.643 KB/s)
397.299 secs (82.477 KB/s)
53.040 secs (617.801 KB/s)
52.844 secs (620.095 KB/s)
441.407 secs (74.235 KB/s)
5.186 secs (6318.790 KB/s)
I forgot to mention: All tests were performed using blowfish. When I changed putty's default settings from AES to blowfish, I noticed no significant difference in performace. I was unable to determine from pscp logs whether or not AES or blowfish was actually being used.