Mailing List Archive

Dejan Macesic wrote:
>
> I'm happy heartbeat works for you Tilman. Unfortunately for me, since
> I'm using a hosting company, adding additional hardware to my machines
> is somewhat of a pain. So I guess I'll continue using fake :)
>

To be fare though. This is an argument against using haertbeats, but not
against using LVS. I don't see why LVS wouldn't work with fake. Maybe I am
missign something or I have something backwards?

Running some sort of "fake"ish thing on a cluster of machines running LVS
naively dsitrbuting across themselves runnin mod_backhand to fix the mistakes
seems like it would work fine -- as convoluted as it is.

LVS will be able to handle more traffic than mod_backhand any day of the week
-- it happens in the kernel. mod_backhand requires the context switching of a
process and proxying the request+response on the application level -- same
overhead (sans the connect() every time) as mod_proxy.

If you can at least get a decent distribution of requests across you machines,
you will gain a considerable advantage. Many people use multiple DNS RR
records to do this, but it has horrible HA problems.

The only time you won't want to do something like this is when ytou are
balancing based on user sessions. This is because, assuming a random
assignment by LVS or some other load balancer, you will be assigned to he
wrong server 1-(1/n) of the time!

Just my two cents.

--
Theo Schlossnagle
1024D/A8EBCF8F/13BD 8C08 6BE2 629A 527E 2DC2 72C2 AD05 A8EB CF8F
2047R/33131B65/71 F7 95 64 49 76 5D BA 3D 90 B9 9F BE 27 24 E7

Correct me if I'm wrong, but I have already downloaded the LVS code and
went through the documentation. LVS requires you use their DNS
package... I couldn't find an example where an external DNS could be
used.

Also, Theo, you mentionned that LVS machines can't also be used to
service web requests. Why so?

Dejan


"Theo E. Schlossnagle" wrote:
>
> Dejan Macesic wrote:
> >
> > I'm happy heartbeat works for you Tilman. Unfortunately for me, since
> > I'm using a hosting company, adding additional hardware to my machines
> > is somewhat of a pain. So I guess I'll continue using fake :)
> >
>
> To be fare though. This is an argument against using haertbeats, but not
> against using LVS. I don't see why LVS wouldn't work with fake. Maybe I am
> missign something or I have something backwards?
>
> Running some sort of "fake"ish thing on a cluster of machines running LVS
> naively dsitrbuting across themselves runnin mod_backhand to fix the mistakes
> seems like it would work fine -- as convoluted as it is.
>
> LVS will be able to handle more traffic than mod_backhand any day of the week
> -- it happens in the kernel. mod_backhand requires the context switching of a
> process and proxying the request+response on the application level -- same
> overhead (sans the connect() every time) as mod_proxy.
>
> If you can at least get a decent distribution of requests across you machines,
> you will gain a considerable advantage. Many people use multiple DNS RR
> records to do this, but it has horrible HA problems.
>
> The only time you won't want to do something like this is when ytou are
> balancing based on user sessions. This is because, assuming a random
> assignment by LVS or some other load balancer, you will be assigned to he
> wrong server 1-(1/n) of the time!
>
> Just my two cents.
>
> --
> Theo Schlossnagle
> 1024D/A8EBCF8F/13BD 8C08 6BE2 629A 527E 2DC2 72C2 AD05 A8EB CF8F
> 2047R/33131B65/71 F7 95 64 49 76 5D BA 3D 90 B9 9F BE 27 24 E7

Dejan Macesic wrote:
>
> Correct me if I'm wrong, but I have already downloaded the LVS code and
> went through the documentation. LVS requires you use their DNS
> package... I couldn't find an example where an external DNS could be
> used.

Here is how I understand it, though I have never really deployed LVS :-)

LVS just uses Linux's IPVS kernel feature to present an VIP (virtual IP) to
the rest of the world and internally NAT it to one of several back end
machines. It has a handful of algorithms that it can use to decide which
server to direct the TCP/IP session to.

> Also, Theo, you mentionned that LVS machines can't also be used to
> service web requests. Why so?

I was corrected on this issue. Someone said that with two interfaces in a
machine (perhaps even virtual) you can use LVS to direct to the local machine.

--
Theo Schlossnagle
1024D/A8EBCF8F/13BD 8C08 6BE2 629A 527E 2DC2 72C2 AD05 A8EB CF8F
2047R/33131B65/71 F7 95 64 49 76 5D BA 3D 90 B9 9F BE 27 24 E7