On Sep 17, 2009, at 5:49 AM, John M. wrote:
It certainly does, thanks! Could I trouble you to explain a little
more about your use of Wackamole and Spread? I’ve not used either of
them before.
There is a How-to here:
You are just using nginx instead of HAProxy, but the Wackamole and
Spread portion still applies.
Scalable Internet Architectures (
http://www.amazon.com/Scalable-Internet-Architectures-Theo-Schlossnagle/dp/067232699X
) also has a section on how this works.
Also, is there any reason why a hosting company would have problems
with such a setup (i.e., this won’t be running in our hardware on
our premises, but we have full control of Linux servers).
Yes, you have to be a little careful here and ask questions up front.
A lot of hosting companies segment their switches such that each port
is it’s own VLAN which means you can’t “float” IPs between ports which
is what you need for this to work. If you tell your hosting company
what you are trying to do and tell them that you need to be able to
have IPs which are programmatically moved between switch ports they
should be able to tell you if this is possible or not. Some hosts may
require you have some sort of “private rack” or other upgrade to make
this possible.
Barry
My experiences with spread were less than stellar, but instead of
going into that, I’ll just give a piece of advice. Spread first tries
to communicate using multicast, and then falls back to broadcasting.
At my hosting provider, since their equipment didn’t support
multicast, this meant that, even though communications were only going
between two computers and did not need to be broadcast to everyone,
all communications were being broadcast to everyone on the subnet. It
didn’t take long before my hosting provider null routed my server. You
can override this behaviour by telling spread to communicate using
unicast, but this only works if there is only one destination for each
source piece of information.
Just something to keep in mind
-Gabe
On Monday, September 21, 2009 at 23:00:33, David M. wrote:
DM> Once again Gena you missed my point by a landslide. WRT/DSL were to
show
DM> proof it could be done and very low end hardware, you could always
get
DM> better hardware.
high traffic load balancer couldn’t be done at WRT/DSL.
original question was about high traffic load balancer.
this is reason, why games with WRT/DSL
are just wasting of his time and money.
experience with WRT/DSL not need to use
nginx on better hardware and better OS.
WRT and DSL - has limited capabilities.
just warning from me, nothing personal.
DM> Furthermore you fail-over logic is flawed, if done proper
DM> you can prevent the failure pages to the end user.
for future new connections - yes, can prevent,
for active tcp sessions/transmissions - can’t.
because tcp connection states and http states are not shared
between independent nodes of nginx-based load balancer cluster.
each hardware node of nginx-based load balancer cluster
may have several hundred or thousand active tcp connections.
this is reason why persistent failure of one node
don’t have zero cost even if other nodes are live.
Another problem with the floating ip is locking arp. The routers on my
host lock the arp for a given ip to whichever mac address it first
hears claiming to have that ip, so I can’t switch ips on the same
segment between machines without talking to them first (or presumably
letting the arp entry expire)
For that you would likely want the DC to setup HSRP so you would have
port
fail over, which would allow for a re-arp, but preventing a “arpstorm”
David
On Thu, Sep 24, 2009 at 8:46 AM, Gabriel R. [email protected]
wrote:
source piece of information.
if this is possible or not. Some hosts may require you have some sort of
why not just ask for your own private vlan? a private vlan will not
only create a boundry around your unciast/broadcast traffic but it
will also allow you to have your own ip unshared ip space (as appose
to shared vlan/shared ip space). Also, private vlan will give you the
frameworkf or moving your ip space anywhere you want inside the
network.
In regards to floating ip, just hava them provision you on a layer2
segment, that will allow you to have multiple ports on their netowrk,
in the same private vlan, in different locations
On Sep 24, Barry A. wrote:
You are just using nginx instead of HAProxy, but the Wackamole and Spread
portion still applies.
How about using one of the LVS solutions? The problem I find with
wackamole is that it assumes the host is “ok” if the network is
reachable. I’d rather have the heartbeat check work off something more
concrete like nginx being up and being able to serve a pre defined
static page.