Re: GBit performance problem with nfs client
I have a file server running on Sarge AMD64 connected
with a 1GBit interface to a GBit uplink off the switch.
Do not think that this sounds like a common problem. It isn't!!!
...
The most interesting fact is, that I obtain about 10MB/s with
my Debian client if I connect the file server to a 100MBit Port
on the switch.
What gbit NIC, which driver, which driver version?
Server:
- Linux SERVER 2.6.8-11-amd64-generic
- BCM95780 with tg3 v3.43f (Jan 9, 2006)
Client Suse:
- Linux SUSE-CLIENT 2.6.8-24.19-default
- Yukon Gigabit Ethernet with default drivers (Suse 9.2)
Client Ubuntu:
- Linux UBUNTU-CLIENT 2.6.15-23-686
- BCM5752KFBG with default tg3 v3.47 (Dec 28, 2005)
Problematic Debian-Client (structurally identical to Suse-Client):
- Linux DEBIAN-CLIENT 2.6.15.1
- Yukon Gigabit Ethernet with default drivers (Debian Sarge AMD64)
Does the driver exibit
different ping times (especially on *variance*) when in 1GB when compared to
100MB?
I am not really sure what "variance" ping time means. Here, my results.
Ping to the GBit connection:
PING SERVER 56(84) bytes of data.
64 bytes from SERVER: icmp_seq=1 ttl=64 time=0.129 ms
64 bytes from SERVER: icmp_seq=2 ttl=64 time=0.095 ms
64 bytes from SERVER: icmp_seq=3 ttl=64 time=0.092 ms
64 bytes from SERVER: icmp_seq=4 ttl=64 time=0.089 ms
64 bytes from SERVER: icmp_seq=5 ttl=64 time=0.093 ms
64 bytes from SERVER: icmp_seq=6 ttl=64 time=0.091 ms
--- SERVER ping statistics ---
6 packets transmitted, 6 received, 0% packet loss, time 5003ms
rtt min/avg/max/mdev = 0.089/0.098/0.129/0.015 ms
Ping to the 100MBit connection:
PING SERVER 56(84) bytes of data.
64 bytes from SERVER: icmp_seq=1 ttl=64 time=0.105 ms
64 bytes from SERVER: icmp_seq=2 ttl=64 time=0.109 ms
64 bytes from SERVER: icmp_seq=3 ttl=64 time=0.104 ms
64 bytes from SERVER: icmp_seq=4 ttl=64 time=0.110 ms
64 bytes from SERVER: icmp_seq=5 ttl=64 time=0.112 ms
64 bytes from SERVER: icmp_seq=6 ttl=64 time=0.113 ms
--- SRVER ping statistics ---
6 packets transmitted, 6 received, 0% packet loss, time 5005ms
rtt min/avg/max/mdev = 0.104/0.108/0.113/0.013 ms
Is the entire *electrical path* from the GB nic to the switch
validated to be CAT5e?
Yes, especially Cat6.
Did you try different cabling?
In an indirect way: Two identical servers, identical OS, and client
tested with different cables.
Greetings,
Gordon
Reply to: