|
Hi,
I deployed an LVS-NAT based login farm at my site a while back, and I
have noticed horrible network congestion issues when people are using
the realservers. NFS glitches are endemic, and cause some problems with
some of our applications. Lots of UDP packet loss and so on. I was
wondering if this might be related to the network topology -- I am using
a one nic, one network NAT setup, with the director as the default
gateway for the realservers. The director is a 666Mhz PIII PC with 128MB
RDRAM and a 10/100BTx interface and the realservers are all HP B132s
with 10BT NICs. I'm planning on replacing the director's NIC with a
better quality 3COM 10/100BTx card at some point to see if that eases
the problem. The realservers are all running HP-UX 10.20 (something that
I cannot change, at least for now).
Anyway, I thought that if I changed the LVS mechanism to DR, I might be
able to eliminate this network performance problem (none of the other
(older, slower) systems currently used for X sessions and telnet access
demonstrate this problem). Unfortunately, there does not appear to be an
easy way to handle the ARP problem on the HP-UX 10.20 realservers. I did
come across Roberto Nibali's post
(http://marc.theaimsgroup.com/?l=linux-virtual-server&m=98741108701884&w=2)
regarding an LVS performance evaluation which hints at a solution for
HP-UX, and I was wondering if somebody might be able to enlighten me.
Or maybe I'll just have to get me a gigabit nic for the director.
Regards,
--
Malcolm Cowe.
IT | Technical Computing, Telephone: +44 131 331 6466
Agilent Technologies Ltd. Telnet: 313-3466
|
