carlsson
Veteran Member
While the following is non-vintage, it may to some extent apply to networking vintage computers too, I'm afraid. Anyway, the scenario:
1. Debian Linux server with integrated VIA Rhine 10/100 ethernet.
2. Netgear 10/100 router w/ four ports plus wireless
3. 3com SuperStack 10 Mbps switch
4. various Windows PCs with 10/100 network cards, different kinds
5. external shell account for testing from outside
All devices were set to autodetect, which I know is a bad thing when you have a heterogene network where two devices neccessarily don't autosense in the same way. Anyway, both the server and Windows PCs claimed to run 100 Mbps full duplex, at least those connected directly through the router.
File transfers, both via local network/Samba and ftp were tremendously slow, maybe like a 56 kps modem or so. To move a file greater than 4 MB from or to the server probably would be faster to burn a CD and install it from there.
First I tried to disconnect the 10 Mbps switch, in case it interferred, but no difference. Then I swapped ports in the router, in case one particular port had been locked by hardware. Still no difference. Then I by force had the server's network card to go down to 10 Mbps half duplex, while the rest of the setup was unchanged. Voila! Now I'm getting reasonable speed, at least for a 10 Mbps connection.
Next week I'll try another router to see if I get any different behavior. Of course the error could be in the server, but I don't see why. Since the external connection to Internet is in the 10 Mbps class anyway, I guess I don't take a performance hit by running the server at the same speed.
Previously I have experienced similar things on older PCs, but those were with a third party installed network card and I found that certain combos of motherboard chipset and NIC worked better than other (so by swapping performance wise identical cards I would rise the transfer speed on two machines). This is an ASRock socket A mobo with integrated controller. Maybe it is the driver inside Linux which could benefit from being updated or replaced?
1. Debian Linux server with integrated VIA Rhine 10/100 ethernet.
2. Netgear 10/100 router w/ four ports plus wireless
3. 3com SuperStack 10 Mbps switch
4. various Windows PCs with 10/100 network cards, different kinds
5. external shell account for testing from outside
All devices were set to autodetect, which I know is a bad thing when you have a heterogene network where two devices neccessarily don't autosense in the same way. Anyway, both the server and Windows PCs claimed to run 100 Mbps full duplex, at least those connected directly through the router.
File transfers, both via local network/Samba and ftp were tremendously slow, maybe like a 56 kps modem or so. To move a file greater than 4 MB from or to the server probably would be faster to burn a CD and install it from there.
First I tried to disconnect the 10 Mbps switch, in case it interferred, but no difference. Then I swapped ports in the router, in case one particular port had been locked by hardware. Still no difference. Then I by force had the server's network card to go down to 10 Mbps half duplex, while the rest of the setup was unchanged. Voila! Now I'm getting reasonable speed, at least for a 10 Mbps connection.
Next week I'll try another router to see if I get any different behavior. Of course the error could be in the server, but I don't see why. Since the external connection to Internet is in the 10 Mbps class anyway, I guess I don't take a performance hit by running the server at the same speed.
Previously I have experienced similar things on older PCs, but those were with a third party installed network card and I found that certain combos of motherboard chipset and NIC worked better than other (so by swapping performance wise identical cards I would rise the transfer speed on two machines). This is an ASRock socket A mobo with integrated controller. Maybe it is the driver inside Linux which could benefit from being updated or replaced?