jim@applix.UUCP (Jim Morton) (03/23/89)
Has anyone else encountered NFS hangs while doing writes from an AIX RT to a Sun disk partition? Near as I can tell, the r_inetlen parameter specified in the /etc/net file can not be 1518 when you are doing NFS mounts to Suns...although it works fine to Ultrix 2.X VAX machines. I can't find any documentation on this "maximum packet length - r_inetlen" paramter and am wondering if 1518 is "standard" in the rest of the TCP/IP world or what...why does IBM ship 1518 as the default inetlen if a popular platform like Suns can't handle packets over 1500?? (At 1518, the Sun console would print "ie0: huge packet") From Sun: /usr/include/net/if_ieee802.h:#define MAX_8023_DLEN 1500 -- Jim Morton, APPLiX Inc., Westboro, MA UUCP: ...harvard!m2c!applix!jim jim@applix.m2c.org
guy@auspex.UUCP (Guy Harris) (03/27/89)
>I can't find any documentation on this "maximum packet length - >r_inetlen" paramter and am wondering if 1518 is "standard" >in the rest of the TCP/IP world or what...why does IBM ship 1518 as >the default inetlen if a popular platform like Suns can't handle packets >over 1500?? More to the point, why does IBM ship 1518 as the default "inetlen" - assuming "inetlen" is the maximum size of an IP datagram - if the DEC/Intel/Xerox Ethernet 2.0 spec says, quite clearly, that an Ethernet packet must have between 46 and 1500 octets? A limit of 1500 wasn't Sun's idea.... The part that's somewhat suspicious here is that the largest valid *Ethernet* packet is 1518 octets long - 6 octets of destination address, 6 octests of source address, 2 octets of type, 1500 octets of data, and 4 octets of frame check sequence. If "inetlen" is really the maximum size of an IP datagram, perhaps somebody at IBM was confused and thought 1518 was the maximum number of *data* bytes in an Ethernet frame, rather than the maximum *total* number of bytes in an Ethernet frame? >(At 1518, the Sun console would print "ie0: huge packet") Yeah, by the Ethernet 2.0 spec, it *is* huge - or, at least, larger than is considered valid - if it has 1518 bytes of data, and thus 1536 bytes in the frame as a whole.... >From Sun: >/usr/include/net/if_ieee802.h:#define MAX_8023_DLEN 1500 Well, that's IEEE 802.3, not Ethernet; however, as I parse stuff in the May 7, 1982 IEEE P802 draft, the recommended MaxFrameSize for a 10Mbit/second CSMA/CD network (for which read "the IEEE 802 version of Ethernet", basically) is 1518, which means the maximum data size in the packet is 1500.