What is the best way to scale ATS?

[Sean G]

Hi Everyone,

Second post, so still very new (but loving it!) to ATS. I see that
clustering support was removed in ATS 7 and I'm wondering how people now
scale ATS ?

I have a single instance at the moment which is likely to hit its limits on
bandwidth before anything else. Is it possible to horizontally scale ATS or
is it best to use transitional load balances in front of ATS? I'm mainly
interested in scaling up resources (bandwidth) but also interested in
resilience as aside benefit.

Many thanks.



--
Sent from: http://apache-traffic-server.24303.n7.nabble.com/

[Susan Hinrichs]

We use a load balancer in front of our clusters of ATS servers to scale
horizontally. You could also just leverage ECMP routing in your router to
similarly distribute connections (plus some extra work to deal with servers
coming up and down). We have a proprietary CARP plugin to share cached
data between servers. Moving forward I think that logic is being folded
into the next hop logic, parent selection logic.

Post by Sean G
Hi Everyone,
Second post, so still very new (but loving it!) to ATS. I see that
clustering support was removed in ATS 7 and I'm wondering how people now
scale ATS ?
I have a single instance at the moment which is likely to hit its limits on
bandwidth before anything else. Is it possible to horizontally scale ATS or
is it best to use transitional load balances in front of ATS? I'm mainly
interested in scaling up resources (bandwidth) but also interested in
resilience as aside benefit.
Many thanks.
--
Sent from: http://apache-traffic-server.24303.n7.nabble.com/

[Leif Hedstrom]

We use a load balancer in front of our clusters of ATS servers to scale horizontally. You could also just leverage ECMP routing in your router to similarly distribute connections (plus some extra work to deal with servers coming up and down). We have a proprietary CARP plugin to share cached data between servers. Moving forward I think that logic is being folded into the next hop logic, parent selection logic.

In addition, for a small number of servers, multiple A-records and some sort of IP-take over would probably work well (using e.g. the other “CARP” :-). Funny enough, the CARP that Susan refers to is Microsoft’s Cache Array Routing Protocol (or some such), whereas traditional CARP is

https://en.wikipedia.org/wiki/Common_Address_Redundancy_Protocol


For a small set of servers, I’d look at something like that I think. Another known to work option is Linux’s HA features, like Heartbeat and Pacemaker (I think).

Cheers,

— leif

[Sean G]

Thanks. I'll take a look.





--
Sent from: http://apache-traffic-server.24303.n7.nabble.com/