|
> Lately the real servers are all running under heavier than
> normal loads without any increase in http requests. Could
> the LVS be getting 'tired' ? I've downloaded netparse and
According to old 'Transformers' episodes your LVS director might be running
low on energon. OK OK -- seriously -- no, your LVS director won't get
tired. However, it might be getting overworked from routing too much
traffic through NAT mode. How much bandwidth is it routing at peak times,
and more specifically how many packets per second?
When you say the load is going up on your real servers, what is actually
happening? What part of the system is actually causing the load to go up?
FYI 2.2.17 is a pretty old kernel, you should update to 2.4.x if possible.
The LVS-NAT is supposed to be much quicker there, and SMP is better. At the
very least update to the latest 2.2.x kernel; 2.2.17 has many security
holes.
> testlvs but the documentation is so scant that I've not been
> able to use them. How can I test to see if the LVS is
> becoming a bottleneck?
Use testlvs before your production setup is production or try it out on the
same machine hardware same kernel QA setup. There's a few examples of how
to use testlvs in the included documentation or in the archives.
> Does it matter that the LVS director here is slower than most
> of the real servers it's directing?
LVS hardware in NAT mode needs to be determined separately than your real
servers. The important figure is going to be how much bandwidth &
packets-per-second you are doing. E.g., for some people that means their
application servers BIG-CGI-SCRIPT capable of doing 1mbit/sec each with dual
P3-1ghz machines can be served by a pentium 200mhz or possibly even less
;-).
P
|
