|
Hello Don,
> Uh. Yea I would think so. Would that be 6x4port or 3x8port NICS? Who is doing
It'll be 6x4 Quadboard NICs and ratz's make-things-work-correctly-although-
alan-cox-does-not-believe-it patch to support such a lot of interfaces in the
linux kernel. Do 8 port NICs really exist that can also be used with COTS hw?
> this? In any case, this would definately be a special case. I was making a
> point that you are fine with NAT for -most- applications.
Ok, our company is doing it and I was the poor guy that had to write the
software for it :)
See, contrary to certain people's belief even bigger companies (2000+
employees) mostly don't have a high traffic website that could saturate
a LVS load balancer. I'm seeing this on a daily basis when I check the
MRTG output for some of our deployed LBs. The biggest issues are the
healthchecks which can bring a LB to its knees if developed in a wrong
way. Application level healthchecking is extremely costly and has the
biggest performance impact on the machine because then actually the
machine is really doing something in user space. But then again, we have
a huge advantage over commercial load balancers: We can run any arbitrary
healtcheck in any programming language we like [yes, even java :)] as
long as it is supported by linux. This makes a LVS load balancer invaluable
for high availability. I don't specifically address you, Don, with this,
but all the guys out here trying to figure out how for heaven's sake they
could find out when they have to take out their webservers because the
backend db couldn't proceed with the request due to a hardware failure
of the securID authentication server, and all this with for example a
CISCO WebNS load balancer.
> > It's a tradeoff between how much time the implementation manager gives
> > you to fiddle around with the arp problem and how saturated the lb will
> > really be when not choosing DR or TUN. Managers tend to use low cost _and_
> > time friendly solutions which don't mix well with the DR approach, when
> > done the first time. I would say: Instead of pissing everyone off trying
> > to set up DR to impress the boss, you'd rather go and impress him with
> > a NAT setup done in 2 hours (with testing).
>
> Yea, unless you have a setup that needs DR. :)
I'm wondering what topology really demands for DR only?
[People tend to generally overestimate resources when it comes to HA issues
and that's why companies such as AC and PWC still get project assignments in
this field]
Best regards,
Roberto Nibali, ratz
ps.: Could you ask Steinberg to port their software to Linux, please 8)
--
mailto: `echo NrOatSz@xxxxxxxxx | sed 's/[NOSPAM]//g'`
|
