home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.protocols.tcp-ip
- Path: sparky!uunet!mcsun!sunic!aun.uninett.no!ugle.unit.no!odda.er.sintef.no!oak
- From: oak@odda.er.sintef.no (Olav Kvittem)
- Subject: How "reliable" is UDP
- Message-ID: <1992Sep2.204556.23149@ugle.unit.no>
- Sender: news@ugle.unit.no (NetNews Administrator)
- Organization: SINTEF, Trondheim, Norway.
- Date: Wed, 2 Sep 92 20:45:56 GMT
- Lines: 46
-
- Hello,
-
- I have run some ttcp tests sending UDP-datagrams to test a 34Mbps/FDDI/Ethernet
- network.
-
- If I run the test on a Sun :
-
- packetsize KB/sek CPU usage Mbps Interface
- 8000 2437 96% 20.0 Ether
- 4000 2051 98% 16.4 Ether
- 1400 1166 98% 9.3 Ether
- 1000 941 97% 7.5 Ether
-
- From this I deduce that many packets does not leave the machine before they are
- dropped. Even with an FDDI interface very few of the packets make it out of the
- machine at 8192 packetsize (size of data for socket write operation).
-
- If I do the same tests on a DECStation Ultrix system it seems to be more "reliable" :
-
- Ethernet :
- 8192 1051 30% 8.0 Ethernet
- 8192 3033 92% 24.2 FDDI
-
- We see that the CPU usage is low keeping the Ethernet UDP at near maximum
- Ethernet speed. That might indicate that the DECsystem have a blocked
- write for UDP sockets while the Sun kernel accepts UDP packets at full speed and
- drops them if buffers fill up ? Perhaps there are no buffers in a UDP-implemntation ?
-
- Does anyone have any explanation or observations for this behaviour ?
-
- Regards
- Olav
-
- FDDI :
- --
- Olav Kvittem : SINTEF
-
- RFC Address Olav.Kvittem@delab.sintef.no
- OR Address C=no;ADMD=" ";PRMD=uninett;
- O=sintef;OU=delab;S=Kvittem;G=Olav
- Postal Address SINTEF DELAB
- N-7034 Trondheim
- Phone +47-7-532586(FAX)
- Phone +47-7-596981
- Description Research Scientist - Communication Networks
-
-