Hi Roman Many thanks for getting those Ethernet capture files. The file sizes are rather large, which means that WireShark takes a relatively long time to apply display filters to the files.
But you have captured the whole iscsi conversation, from the first SYN packet, through the iscsi logon, through to data transfer, with reads and writes. And you have captured the same iscsi conversation at BOTH ends, the Windows initiator, and the OpenSolaris target, so we get two aspects on the same packets. Let's look at the first pair of files: "winserver_jambo_off_tso_off.cap" & "sun_jambo_off.cap" In the SYN packet from the initiator, we can see it specifying a Window Scale of 8 (x256), and the SYN+ACK from the target specifies a Window Scale of 3 (x8), http://en.wikipedia.org/wiki/TCP_window_scale_option So file "winserver_jambo_off_tso_off.cap" is captured on the Windows server with WireShark. Looking at the file with WireShark, we see the 'on the wire' packet size is a maximum of 1514 bytes, for both sent and received packets, which is what you would expect if Jumbo frames were disabled. File "sun_jambo_off.cap" is captured on the OpenSolaris server with snoop. For packets received from the Windows initiator, the 'on the wire' packet size is a maximum of 1514 bytes. But many of the packets being sent by OpenSolaris are showing as having a packets size of 8294 bytes. By comparing the two files side by side (in two instances of WireShark), we can see what is happening. The packets with size 8294 being sent by OpenSolaris have a TCP payload of size 8240. But the Windows initiator receives 6 packets: 5 with size 1514 bytes and 1 with 994 bytes. The corresponding TCP payload size is (5*1460)+940 = 8240. So all the above looks Ok. I'm out of time for now, so i will report back on the second set of capture files later... Regards Nigel Smith -- This message posted from opensolaris.org _______________________________________________ storage-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/storage-discuss
