[c-nsp] Cisco 7600 vs Juniper M7i

[lists at hojmark.org]

>> Can we please have the _full_ ACLs with _all_ terms exactly
>> as used on the DUTs, and all other config too? The test is not
>> repeatable without those.

> I guess the non-response can be considered a "no". I'm drawing
> my conclusions from that.

IMO, the test in itself isn't all that interesting: The M10i is
a 6G platform, so when you send it 6G of traffic as 64B packets
with lots of features turned on, it'll drop packets... It's as
simple as that.

Had the test used IMIX, the M10i wouldn't have missed a beat.
(And that, of cause, is testable if you have the ACLs).

Obviously, the test is constructed that way to make everyone
think Vendor J makes boxes that won't do features without a
performance hit, while in fact they were just testing a specific
platform to it's documented max.

(It's also possible to construct a test where the 7604 will drop
close to 100% when you add features, even when you're very far
from the (close to) 20G you get on the 6724 blade).


What *is* interesting about the test is the level of performance
you get in a 7604 vs a M10i: The first will do close to 20G in
the configuration tested. The second will do close to 6G but is
much more expensive, even when you don't use the expensive PICs,
which were used in the test (but weren't needed for it).

Just my $0.02

-A