>Hi all,
>
>I've got a 10BASE2 which seems to perform abnormally slow. Sending a 50 MB file seems to take about 100 seconds. There are no other workstation causing network traffic, so the bad performance is not due to heavy network trafic.
>
>I would expect that a 10BASE2 should be capable of transmitting 10 Mbps. If I take the overhead of the TCP/IP protocol into account I expect 1 MB per second to be reachable. However if 50 MB/100 Seconds, results in a netto troughput of 500 KB/S.
>
>Can anyone shed a light on this ?
That's normal; 10Base2 (ThinNet), even given proper termination and no collisions, is still fraught with overhead from latency due to the need to listen for 1/2 packet before you can determine that your packet is transmitting cleanly, and as I recall, there's a mandatory delay between consecutive blocks of packets as well. From my notes, and charts from Paul Fortier's "Handbook of LAN Technology, 3rd Edition" provided by Xerox PARC, the people who originated EtherNet, ThinNet is lucky to sustain .5MB/sec, and more realistically, will run in the 300KB/s range with normal heartbeat traffic and a nominal number of nodes on a single segment; multiple segments will introduce additional latency delays on each first packet to permit the broadcast packet to travel inter-segment, and routers/brouters/gateways will introduce even more delay.
Star-wired 10BaseT on a switch offers potentially significant advantages in performance, is cheaper to wire, more robust, and can, if you cable with CAT5 throughtout, normally cand be upgraded in place by replacing NICs with 100BaseTX NICs (typical cost in the $12-$20 range) and the upgrade of the hub (I pay around $8-12/port for cheap 10/100 switches.)
Given that a 10Base2 net will go down if there's a break in the cable anywhere, a loss of termination, jabbering, or induced noise, even small networks are generally better off with 10 or 100BaseT cable plants.
I'd recommend that you do a bit of research on EtherNet; PARC did exhaustive studies of the performance of CSMA/CD architectures implemented as EtherNet back in the early- to mid-80s (MetCalfe et al.) - Andrew Tanenbaum's
Computer Networks, 2nd Edition or later includes some interesting studies of EtherNet, as does the Fortier book mentioned earlier.