On Sun, Apr 06, 2003 at 01:37:55AM +0200, Alex Kramarov wrote:
> Hi.
>
> I am currently running a pair of lvs directors in LVS-NAT mode, handling
> about 10MBit loads. the real servers are NT boxes, that run the dinamic part
> of my site (asp). the rest of the site (static content) is being served by 2
> tux/apache servers, protected by heartbeat. the combined incoming traffic to
> these servers is about 10MBit, but the outgoing is approaching 100MBit, so i
> am about to add another tux/apache server to share the load, and here i will
> have to drop the heartbeat and setup LVS-DR to these 3 servers.
>
> in the lvs howto there is an assumption that a linux box on 100MBit network
> is actually a 8000 pps box - it says this was done in the times of 2.2
> kernels. is this assumption still valid ? summed up, my site currently
> approaching 14000 pps all together in peaks - routing all this traffic
> through a single director in this case could be problematic . the director
> is 2 cpu 1GHZ IBM x330 . is anybody running such setup ? any suggestions ?
>
> i currently have 100MBit switched network on cisco 2900 which is connected
> to the main switch of the provider with several 100mbit ports configured as
> a port-group (etherchannel), so the outbound bandwidth capacity should be
> sufficient for the perpose . it's the pps that troubles me after reading the
> howto. i would really like to stay with 100Mbit nerwork, since none of the
> servers (nor the switch) have Gig interfaces.
>
> At the time i have tested this director to do lvs-nat to all the servers,
> back then the overall traffic of the site was about 30MBit. Of course, i
> will run my own tests, and post them to the list, but if someone has already
> done these, it would really help.
Hi,
I don't exactly follow what the problem is. Is the Linux Director
having trouble keeping up? It certainly shouldn't untill you
approach 1Gbit/s of traffic.
--
Horms
|