We launched the ACX7024 in mid-2022, with the following characteristics:
1RU Ethernet router.
Built for the aggregation of SFP interfaces (1GE/10GE/25GE) to QSFP28 interfaces (40GE/100GE).
360Gbps of forwarding capacity powered by a Broadcom DNX Qumran2u.
Class-D capable based on Juniper's PTP FPGA.
Hardened device, categorized as i-temp (supports temperatures from -40 to +65C).
AC or DC power modules, with 1+1 redundancy.
6x fixed fans (front to back cooling), 5+1 redundancy.
It was targeting markets like metro and aggregation where routing scale is not a key parameter. In these roles, high volumes of routers are deployed and cost pressure is very important.
That's why it was designed with these constraints in mind, and that reflected in the choice of the CPU and the memory size:
Intel ATOM Deveron C3508 : 4 cores @ 1.6GHz, single-threaded
CPU INFO:
========
Model name : Intel(R) Atom(TM) CPU C3508 @ 1.60GHz
Configured CPU Speed : 1600.012 1600.019 1600.005 1600.049 MHz
Physical cores : 4
Hyperthreaded cores : 4
L1 Cache size : 224 kB
L2 Cache size : 8192 kB
L23Cache size :
CPU PAIRS:
=========
Core 0 :
Core 1 : 3
Core 2 : 0
Core 3 :
SYSTEM MEMORY USAGE (MB):
========================
MemTotal: 15995356 kB
Consequently, certain Qumran2u scale "maximum dimensions" can't be reached, the CPU memory being the first bottleneck. Several features can't be enabled
IPFIX
FIB compression
RIB sharding
3rd party App Hosting. Note: PAA (Paragon Active Assurance) is an exception to that rule and is supported on ACX7024.
And this CPU and DRAM are the key differences between ACX7024 and ACX7024X.
The ACX7024 (non-X) port density and form-factor happen to be very attractive for "light multi-service edge router" use-cases, particularly for enterprise and small CPEs. In these network roles, hardware hardening (i-temp) support is not required, the routers being operated in DC controlled environments. In the other hand, the support of features like sampling and potentially full IPv4 and IPv6 internet tables is required.
That's the main motivation behind the introduction of an ACX7024X that only differs from its predecessor at the CPU and DRAM level: same port density and forwarding capability, same ASIC, but a more powerful CPU:
regress@acx7024x-01# set system packet-forwarding-options hw-db-profile ?
Possible completions:
+ apply-groups Groups from which to inherit configuration data
+ apply-groups-except Don't inherit configuration data from these groups
bng Selects BNG DB profile, restarts PFE
carrier-ethernet Selects Carrier Ethernet high scale MAC DB profile, restarts PFE
cloud-metro Selects Cloud-Metro DB profile, restarts PFE
lean-edge Selects Lean Edge max IPv4/IPv6 FIB DB profile, restarts PFE
[edit]
regress@ acx7024x-01#
By default, it will boot up in lean-edge mode and in the example below, the router receives a full internet view from an EBGP peer. That's almost 1 million IPv4 entries and more than 200,000 IPv6.
It's of course an ideal case, since we have a single BGP neighbor. But still you can see the FIB compression in action with a ratio of 83% for IPv4 and 73% for IPv6.
In terms of resource occupation, that's the appid 12 here with 212k IPv4 and 66k IPv6 entries for a total 346k used out of the 1.18M capacity.
The ACX7024X is a very competitive 360Gbps router with 24x SFP, 1GE, 10GE or 25GE and 4 ports QSFP 40GE or 100GE. PTP class-D capable and with a large DRAM space for CPU-intensive features. It supports a full internet table (with both IPv4 and IPv6 routes) and can be used in multiple enterprise roles or for small-size CPEs.
