Chris Robertson wrote:
> Anthony DeMatteis wrote:
>> Hi group,
>>
>> Has anyone played with limiting bandwidth via squid? I currently have
>> three bandwidth managers between my border router and customer base.
>> These servers limit traffic based on service customer has selected
>> (1.5M/768k, etc). There are a number of issues with this solution;
>> interface loads, load balancing, etc. I have implemented a single squid
>> server using wccpv2. While I see this as being beneficial to reduce
>> traffic on my routers upstream interfaces, I don't see it positively
>> affecting the traffic on my bandwidth managers. Traffic will still be
>> passing through these interfaces. In fact, I will have additional
>> traffic as port 80 traffic is diverted to the caching server(s). As I
>> understand wccp, it will load balance between the cache engines that are
>> visible to it. And I'm trying to understand squid pools. I'm thinking
>> I may be able to set up several squid servers using pools to load
>> balance and limit my traffic. Conceptually I envision this:
>>
>> Upstream Provider
>> | |
>> | |
>> Router (wccpv2 - GRE Tunnels to each Caching Server
>> | |
>> | |
>> | |
>> Switch <-Backbone-> Network/Customer Base
>> | | | |
>> | | | - Cache1
>> | | |- Cache2
>> | |- Cache3
>> |- Cache4
>>
>> Cache Servers would keep a replicated database of customer bandwidth
>> limits for use by local squid process. Am I reaching here, or is this
>> within the realm of possibility?
>>
>
> Currently delay pool information is not shared between servers, or even
> between restarts of a single server. But using a source hash algorithm
> on the load balancer would get you in the ball park.
>
>> Any feedback is welcomed. Thank you.
>>
>> Tony DeMatteis
>>
>
> Chris
You may want to consider an architecture using both delay pools and QoS.
So squid performs the front-line bandwidth management for HTTP and marks
it's outbound packets so the other managers can be less lenient or
routed around. The squid are effectively stateless as mentioned b Chris
though.
The QoS settings in Squid are tcp_outgoing_tos (ACL-controlled marks)
and more recently either zph_* (2.7) or qos_flows (3.1) for
HIT/MISS/peer marked traffic.
Amos
-- Please be using Current Stable Squid 2.7.STABLE6 or 3.0.STABLE13 Current Beta Squid 3.1.0.5Received on Fri Feb 06 2009 - 12:16:13 MST
This archive was generated by hypermail 2.2.0 : Fri Feb 06 2009 - 12:00:02 MST