[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: pf address pools
Ok, I remember round-robin DNS, but if you ever had the need to change
DNS servers, and you then see what T-Online, AOL and other "ISP's" do with
settings, you begin to ask if this really works, despite the fact, that you
normally to some
sort of caching for the DNS queries, and there is no guaranty, that your ISP
DNS will ask a
second time. That is the reason why my company still sticks on BigIP.
----- Original Message -----
From: "Dries Schellekens" <[email protected]>
To: "Stefan Sonnenberg-Carstens" <[email protected]>
Cc: <[email protected]>
Sent: Friday, November 29, 2002 1:40 PM
Subject: Re: pf address pools
> On Fri, 29 Nov 2002, Stefan Sonnenberg-Carstens wrote:
> > So, do you think it might be better to use ipfilter than pf on OpenBSD
> > that case ?
> This feature ("round-robin sticky") is not in ipfilter 3.4.30 (released
> this week), so it's only available in ipfilter 4.0 alpha.
> To implement sticky balancing across multiple subnets you have to remember
> which address was used for which client source address. The cost of this
> is memory and time to search through it.
> Calculating a hash is much easier, but you don't have the flexiblity of
> redirecting across multiple subnets.
> > And the next question is, is it useful to have a wide spread (more than
> > IP subnet) servers to do load-balancing on ?
> > After all, that is a feature, the BigIP supports and I know that
> > atleast www.heisse.de is using this, to implement complete redundancy
> > by location seperated servers.
> Then where would you put your load balancing machine? Round-robin DNS is
> much easier to acomplish this.
> I think most other solutions want to servers in the same subnet.
> For instance if you want to do "Virtual Server via Direct Routing" (linux
> virtual server project calls it this way), the servers need to be on the
> same subnet: http://www.linuxvirtualserver.org/VS-DRouting.html
> BTW this technique is much better than a Virtual Server via NAT, because
> only the request needs to pass through the load balancer, while the reply
> does not.
> > ----- Original Message -----
> > From: "Darren Reed" <[email protected]>
> > To: "Jedi/Sector One" <[email protected]>
> > Cc: <[email protected]>
> > Sent: Friday, November 29, 2002 9:53 AM
> > Subject: Re: pf address pools
> > > In some mail from Jedi/Sector One, sie said:
> > > >
> > > > On Thu, Nov 28, 2002 at 11:59:37PM +0000, Ryan McBride wrote:
> > > > > rdr on $ext_if from any to $public_ip port 80 -> \
> > > > > 192.168.0.4/30 source-hash
> > > >
> > > > As a side note, source-hash (a feature called 'sticky balancing'
> > > > hardware load balancers) is very useful for web servers with PHP
> > > >
> > > > - by default, PHP save sessions in local files.
> > > > - to speed up things, it's also possible to use shared memory.
> > > > - poorly written PHP scripts (those that customers like to install)
> > > > create temporary files in /tmp.
> > > >
> > > > Without sticky balancing, a typical syndrom is that users have to
> > > > re-authenticate several times while browsing a web site.
> > >
> > > Well I don't think the above is a good implementation of sticky
> > > load balancing because it confines your destination IP addresses
> > > to be a single subnet mask range.
> > >
> > > I did sticky redirection for IPFilter last month, I think, and that
> > > implementation does not have this problem. More importantly, the
> > > stickiness can be mixed with any other redirection options.
> > >
> > > If routers use the above for stickiness then said routers suck, IMHO.
> > >
> > > Darren
> Dries Schellekens
> email: [email protected]