allow me to shed some needed light on this thread.
current deal at sector for networking stuff:
-ATT adsl2+ (not vdsl2, the loop is too long, would do no good, and
ATT has no VRAD chassis anywhere near the site anyway)
-11.2 mbit usable phy rate downstream
-1.1 mbit usable phy rate upstream
-shaped in cisco 7505 router to 10.5 mbit downstream, and .98 mbit upstream
-shaped queues have the following QoS policies applied:
policy-map downstream
class class-default
fair-queue
fair-queue queue-limit 32
policy-map upstream
class class-default
fair-queue
fair-queue queue-limit 8
-the router has plenty of memory for hundreds of thousands of NAT
flows/dynamic translations
-the 'core switch' doing layer-3 vlan boundary duty is stateless and
can handle any collection of ethernet/IP traffic thrown at it
the "qos" created with fair-queue is fine for then few ten's of dozens
that ever compete for up/downstream bandwidth -- the issue that may be
encountered hosting a media related site (even if only for small
files) would be, naturally, content and popularity dependent.
at the outside, the service times/reply times would be exaggerated (by
upstream contention) to the point where remote users might just give
up.
i might suggest that sector pool some resources to toss a 1u box in
the 5nines sandbox, for a low rate, where we can hand out gige ports
without a lot of hassel, and where we have multiple gige's of capacity
on tap.
-tk
> --
>
>