|
One of the commercial implementation based on LVS kernel
is http://www.redhillnetworks.com . They hired an independent
lab tested the product based on LVS kernel ( all user space
code was proprietary) and found that at close to your
configuration, their box can take 4000 new connection/sec
and 118,000 concurrent connections. Of course, they
fine tuned the kernel little bit, but all the LVS code was
not changed. Hope this will give you some idea how
wonderful LVS is!
At 11:09 AM 7/12/00 -0400, Kirk Bauer wrote:
>As a long time linux user and developer, I'm glad to see an open
>source alternative to the Cisco localdirector and such.
>Thanks for all of your hard work.
>
>In any case, I plan to use LVS to provide load balancing and
>high-availability for my company.
>
>Our initial setup will be 2 load balancers with 2 real servers. Each
>real server will be running a web server as well as a custom
>application.
>
>The custom application listens on a socket that every user of our
>product connects to. Each user maintains one open TCP connection
>throughout the duration of their use of our product. I would say that
>the load balancing of our custom application will be much like the load
>balancing of say SSH. The average connection time will be much greater
>than web traffic... and the total number of connections will be much
>greater at any given time.
>
>I am trying to design a system now that will scale to the following
>numbers:
> Web traffic: Around 500 hits/second
> Application traffic: Support up to 20,000 simultaneous connections
>
>It seems to me that the easiest setup is VS-NAT. I can see several
>benefits with this setup... namely that the real servers will be
>firewalled, and there is no special setup required on them.
>
>However, my concern is whether NAT can support the number of
>simultaneous connections I mentioned above. We are planning on using a
>Celeron 500 with 128MB of RAM for the load balancing machines.
>
>So, my question to you is - can I use VS-NAT and support the mentioned
>number of connections? If I remember correctly, I'll have to increase
>some constants when compiling VS to support that number of
>connections... but will masquerading be able to handle it? How many
>simultaneous connections can one VS-NAT machine possibly handle?
>
>Right now, traffic throughput isn't much of a problem, as we only have a
>1Mbit pipe to the internet. However, it was mentioned that this machine
>should be able to masquerade at nearly the full 100Mbps speed with the given
>processor (and two 100Mbps cards).
>
>My other alternative is to use VS-DR... in which case I would still have
>to make sure that VS itself could handle that many simultaneous
>connections.
>
>--
>Kirk Bauer <kirk@xxxxxxxxxxxxxxx> (404) 964-6071
> Chief Technology Officer, TogetherWeb Inc
>
>
>
|
