Various tests using Netpipe 2.3 showing network throughput as a function of packet size. Performed on 5/09 - 5/11,5/16 2000, by David Mathog (mathog@caltech.edu). Some DS10RH62 tests required Netpipe 2.4beta, these are inicated with a * (one test was repeated with both programs and results were identical). The tests show throughput as a function of the number of bytes in a test buffer transferred. File: is Beta Details nptcp.exe exe Netpipe 2.3 compiled on 7.2-1 for for Alpha systems. netpipe.obj obj ditto tcp.obj obj ditto DS10RH62DS10RH62.JPG plot DS10/Linux RH6.2 ewa0 100baseT -> Linksys switch -> DS10/Linux RH6.2 ewa0 100baseT DS10RH62RH52INTEL.JPG plot * DS10/Linux RH6.2 ewa0 100baseT -> Cisco switch -> RH 5.2 Intel 100BaseT DS10RH62SELF.JPG plot DS10/Linux RH6.2 ewa0 100baseT -> localhost RH52INTELSELF.JPG plot RH 5.2 Intel 100BaseT -> localhost DS10multinetDS10RH62.JPG plot * DS10/VMS Multinet 4.2A-X ewa0 100baseT -> Linksys switch -> DS10/Linux RH 6.2 ewa0 100baseT DS10multinetRH52INTEL.JPG plot DS10/VMS Multinet 4.2A-X ewa0 100baseT -> Cisco switch -> RH 5.2 Intel 100BaseT DS10multinetSELF.JPG plot DS10/VMS Multinet 4.2A-X ewa0 100baseT -> localhost DS10ucxDS10rh62.JPG plot * DS10/VMS TCPIP 5.0 ewa0 100baseT -> Linksys switch -> DS10/Linux RH 6.2 ewa0 100baseT DS10ucxDS10multinet.JPG plot DS10/VMS TCPIP 5.0 ewa0 100baseT -> Linksys switch -> DS10/VMS Multinet 4.2A-X ewa0 100baseT DS10ucxRH52INTEL.JPG plot DS10/VMS TCPIP 5.0 ewa0 100baseT -> Cisco switch -> RH 5.2 Intel 100BaseT DS10multinetDS10ucx.JPG plot DS10/Linux RH 6.2 ewa0 100baseT -> Linksys switch -> DS10/VMS TCPIP 5.0 ewa0 100baseT CTRU64CTRU64.JPG plot Clone 164 Tru64 -> 3COM hub -> Clone 164 Tru64 CTRU64SELF.JPG plot Clone 164 Tru64 -> localhost COMPARISON.GIF plot many of the above in one plot DS10MULTINET.GIF plot all DS10 Multinet plots DS10UCX.GIF plot all DS10 TCP/IP services plots DS10RH62.GIF plot all DS10 RH 6.2 plots RH52INTEL.GIF plot all RH 5.2 intel plots 23versus24.GIF plot DS10MultinetRH52Intel for Netpipe 2.3 and 2.4Beta DS10RH62DS10RH62.TXT raw DS10/Linux RH6.2 ewa0 100baseT -> Linksys switch -> DS10/Linux RH6.2 ewa0 100baseT DS10RH62RH52INTEL.TXT raw * DS10/Linux RH6.2 ewa0 100baseT -> Cisco switch -> RH 5.2 Intel 100BaseT DS10RH62SELF.TXT raw DS10/Linux RH6.2 ewa0 100baseT -> localhost RH52INTELSELF.JPG raw RH 5.2 Intel 100BaseT -> localhost DS10multinetDS10RH62.JPG raw * DS10/VMS Multinet 4.2A-X ewa0 100baseT -> Linksys switch -> DS10/Linux RH 6.2 ewa0 100baseT DS10multinetRH52INTEL.TXT raw DS10/VMS Multinet 4.2A-X ewa0 100baseT -> Cisco switch -> RH 5.2 Intel 100BaseT DS10multinetSELF.TXT raw DS10/VMS Multinet 4.2A-X ewa0 100baseT -> localhost DS10ucxDS10multinet.TXT raw DS10/VMS TCPIP 5.0 ewa0 100baseT -> Linksys switch -> DS10/VMS Multinet 4.2A-X ewa0 100baseT DS10ucxDS10rh62.TXT raw * DS10/VMS TCPIP 5.0 ewa0 100baseT -> Linksys switch -> DS10/Linux RH 6.2 ewa0 100baseT DS10ucxRH52INTEL.TXT raw DS10/VMS TCPIP 5.0 ewa0 100baseT -> Cisco switch -> RH 5.2 Intel 100BaseT DS10multinetDS10ucx.TXT raw DS10/Linux RH 6.2 ewa0 100baseT -> Linksys switch -> DS10/VMS TCPIP 5.0 ewa0 100baseT CTRU64CTRU64.TXT raw Clone 164 Tru64 -> 3COM hub -> Clone 164 Tru64 CTRU64SELF.TXT raw Clone 164 Tru64 -> localhost Synonyms: (files which I duplicated to make the names more meaningful, leaving the originals around too) DS10multinetRH52INTEL = TORH52INTEL DS10multinetSELF = VMSLOCAL DS10RH62DS10RH62 = TOMONKEY01 DS10multinetRH52INTEL_BETA used Netpipe 2.4beta and DS10multinetRH52INTEL used 2.3. The results indicate only "noise" differences between the tests. The "Clone 164" systems were two identical 21164 clones, but I don't know which motherboard or clockspeed, running Tru64. No tests from the Tru64 sytems to the others were performed because the Tru64 machines were located in a different subnet, with a firewall in between. Tests that failed with Netpipe 2.3 DS10 RH6.2 ewa0 <-> DS10 VMS 7.2-1 Multinet ewb0 DS10 RH6.2 ewa0 <-> DS10 VMS 7.2-1 TCPIP 5.0 ewa0 DS10 RH6.2 ewa0 <-> RH 5.2 Intel These had to be rerun later (5/15/00) using Netpipe 2.4beta. Tests that could not be run: DS10 Multinet <-> DS10 Multinet. Apparently there is some sort of copy protection built into Multinet, so that when I attempted to reuse the PAK for this test on a second system it would not let the two Multinet systems communicate with each other. (The second system could communicate with RH62 and TCPIP 5.0, so it was working in general.) If this isn't a copy protection mechanism it's one hell of an annoying bug! Other notes: DS10 TCPIP services and DS10 RH 6.2 were tested against RH5.2 Intel by running a 50 foot cable from the second ethernet port on each box to a jack on a different subnet. Note that DS10multinetDS10ucx and DS10ucxDS10multinet are almost identical except for what looks like noise. That is, this test appears to be symmetric. Apparently in the awful interaction between Multinet on a DS10 and RH 5.2 Intel one of the many spikes is intrinsic to Redhat, and the rest are due to the interaction between the two stacks. ********************************************************************** Testing details: Netpipe 2.3 is from: ftp://ftp.scl.ameslab.gov/pub/netpipe/netpipe-2.3.tar.gz Code changes needed for OpenVMS. In netpipe.c change PrintUsage() to void PrintUsage(void) and in netpipe.h insert after the last #include #ifdef __VMS #include <unixio.h> #endif Build with: $ mycc:= cc/standard=relaxed/prefix=all/define=TCP $ mycc NETPIPE.C $ mycc TCP.C $ link netpipe,tcp $ nptcp:==$wherever:netpipe.exe Full VMS distribution may be obtained from: http://seqaxp.bio.caltech.edu/pub/SOFTWARE/NETPIPE_2_3.ZIP All receivers ran: NPtcp -r All senders ran: NPtcp -t -h nodename -o testtonodename -P Plotted with: $ gnuplot set logscale set data style lines set terminal postscript set output "testtonodename.ps" plot "testtonodename" using 4:2 exit ********************************************************************** Hardware/Software details Process Software MultiNet V4.2 Rev A-X, COMPAQ AlphaServer DS10 466 MHz, OpenVMS AXP V7.2-1 RH 6.2 on DS10 RH 5.2 on a dual PII 400 Intel, 100 BaseT TCPIP V5.0-10, VMS V7.2-1