Results Summary and Analysis

ACX7024 Functions and Performance

During the validation process, we successfully demonstrated a robust solution for 5G xHaul transport infrastructure using Seamless MPLS with Segment Routing. The JVD achieved a reasonable scale of L2/L3 connectivity services, meeting the expectations of Mobile Network Operators (MNOs) and Metropolitan Area Network (MAN) operators for real network deployments. The solution also met stringent Service Level Agreement (SLA) requirements.

ECMP Load Balancing

The design reduces traffic impact during link/node failure events by enabling load sharing ECMP operations across all devices. Several ECMP mechanisms were configured (as supported) including adjusting IGP metrics, BGP multipath, ECMP fast-reroute, and VPN-unequal-cost for L3VPN services. In addition, FAT-PW label is enabled on the ACX7000 series for L2VPN and L2Circuits. EVPN FAT-PW is supported starting in Junos OS Evolved Release 23.1R1.

For a copy of the full test report, including details on hash-keys enabled for this validation and traffic load sharing limitations, contact your Juniper Networks representative.

Table 1: ACX7024 ECMP Summary Results

ACX7000 ECMP Load Balancing Performance [*]
Service: DUT (Traffic Path)	ECMP Links	Flow	FAT-PW	Link1	Service: DUT (Traffic Path)	ECMP Links	Flow
L2VPN: ACX7024 (AN4 to SAG)	4	100kfps	Y	ae22 20.2kpps	ae25(1) 30.3kpps	et-0/0/5 30.3kpps	et-0/0/6 20.2kpps
L2CKT: ACX7024 (AN4 to SAG)	3	100kfps	Y	ae22 35kpps	et-0/0/5 33kpps	et-0/0/6 32kpps	N/A
VPLS: ACX7024 (AN4 to SAG) [1]	4	100kfps	N	ae22(1) 24.2kpps	ae25(1) 26.7kpps	et-0/0/2 24.2kpps	et-0/0/3 26.7kpps
EVPN: ACX7024 (AN4 to DU)	4	100kfps	N	ae22(1) 24.7kpps	ae25(1) 26.2kpps	et-0/0/2 24.7kpps	et-0/0/3 26.2kpps
L3VPN: ACX7024 (AN4 to AG1.1)	4	100kfps	N	ae22(1) 25.7kpps	ae25(1) 25.9kpps	et-0/0/2 25.7kpps	et-0/0/3 25.9kpps

[*] For complete ECMP results with all outputs, contact your account representative.

[1] Only Known Unicast is shown. VPLS BUM traffic should not load balance over ECMP routed links. Expected behavior.

In terms of ECMP performance, the ACX7024 performed similarly to the previously tested 5G Fronthaul Network Using Seamless MPLS Segment Routing JVD. However, there was a slight imbalance in the distribution of L2VPN traffic due to the hash computation on the ACX7024. Similar results were observed when using three ECMP links, with the ACX7024 exhibiting a distribution of 33kpps/38kpps/30kpps, while the ACX7100-48L achieved nearly perfect balance. For a detailed report on the test results, including information on ACX7024 ECMP Load-Balancing, contact your Juniper Networks representative.

Network Convergence

Overall convergence results are within expectations for the given network design. In Fronthaul (CSR to HSR), ACX7024 failure and restoration events were well within 50ms recovery where expected. ACX7024 performance was comparable to ACX7100-48L in the CSR role with the ACX7100-48L reasonably achieving slightly better convergence results. All ACX7000 series demonstrate improved convergence compared to previous generation ACX5448/ACX710 where CSR-to-DU reported up to four seconds of traffic loss during EVPN-VPWS failure events.

Table 2 summarizes convergence performance validations across all represented VPN services, which includes single-homing or active-active multi-homing. Traffic is sent as known-unicast. Higher convergence can be expected for BUM traffic in MAC-learned services. These results are also recorded in the full test report.

Table 2: Convergence Times for 5G Fronthaul Failure Events Per Flow Type

Flow Type	EVPN-VPWS (msec)		EVPN-FXC (msec)	EVPN-ELAN (msec)	VPLS (msec)	L2VPN (msec)		L2CKT (msec)	L3VPN (msec)
Single/Multihoming	SH	A/A MH	SH	A/A MH	A/A MH	SH	SH		SH
AN4 to AG1.1 disable	18	10	10	11	0	18		21	15
AN4 to AG1.1 enable	2	2	2	0	0	5		2	8
AN4 to AG1.2 disable	31	20	0	20	15	0		5	22
AN3 to AG1.2 enable	0	4	2	2	4	2		4	8

Class of Service Validation

Across the end-to-end topology, classification and rewrite was performed on 802.1p, DSCP, and EXP as outlined in Figure 1 . Table 3 summarizes these results for the included services and classification types. In dual-tag scenarios, the outer service tag is used for classification and rewrite. CoS bits can be preserved end-to-end, including for inner or outer tags.

When a port shaper is defined, applicable class of service functions adjusted to the new port speed and performed equivalently. For example, a 1G port shaper was used and transmit-rate percentages were correctly shown to be based on a 1G port speed.

Table 3 summarizes VLAN operation scenarios that we executed and the corresponding results. VLAN Tags are represented as Untagged (UT), Single-Tagged (ST), and Dual-Tagged (DT). All listed input/output VLAN mapping operations were validated across L2Circuit, L2VPN, EVPN-VPWS, and EVPN-ELAN services.

For the full test report, which includes an analysis explaining the results for each function, contact your Juniper Networks representative.

Table 3: CoS Summarized Results

Traffic Scenario	VLAN	Ingress Classification Mapped to FC			Scheduler Honored		Rates	Codepoints Rewritten				Bits Preserved
Fixed Classifier	TAG	802.1p	DSCP	EXP	SH	LOW		802.1p	DSCP	EXP	E2E
EVPN-VPWS	UT	--	--	√	√	--		√	--	√	√
EVPN-ELAN	UT	--	--	√	√	--		√	--	√	√
L2Circuit	UT	--	--	√	√	--		√	--	√	√
pop / push	DT	--	--	√	√	--		√	--	√	√
swap / swap	DT	--	--	√	√	--		√	--	√	√
swap-swap / swap-swap	DT	--	--	√	√	--		√	--	√	√
pop-swap / swap-push	DT	--	--	√	√	--		√	--	√	√
pop-pop / push-push	DT	--	--	√	√	--		√	--	√	NA
push / pop	ST	--	--	√	√	--		√	--	√	√
swap / swap	ST	--	--	√	√	--		√	--	√	√
pop / push	ST	--	--	√	√	--		√	--	√	√
swap-push / pop-swap	ST	--	--	√	√	--		√	--	√	√
BA Classifier	TAG	802.1p	DSCP	EXP	SH	LOW		802.1p	DSCP	EXP	E2E
L3VPN	UT	--	√	√	--	√		--	√	√	√
L2VPN	UT	√	--	√	--	√		√	--	√	√
BGP-VPLS	UT	√	--	√	--	√		√	--	√	√
pop / push	DT	√	--	√	--	√		√	--	√	√
swap / swap	DT	√	--	√	--	√		√	--	√	√
swap-swap / swap-swap	DT	√	--	√	--	√		√	--	√	√
pop-swap / swap-push	DT	√	--	√	--	√		√	--	√	√
pop-pop / push-push	DT	√	--	√	--	√		√	--	√	NA
push / pop	ST	√	--	√	--	√		√	--	√	√
swap / swap	ST	√	--	√	--	√		√	--	√	√
pop / push	ST	√	--	√	--	√		√	--	√	√
swap-push / pop-swap	ST	√	--	√	--	√		√	--	√	√

Congestion Scenarios

The validation included various congestion scenarios outlined in the Solution Validation Goals section. Congestion constitutes one or more conditions where traffic exceeds the configured scheduler transmit-rate, shaped-rate, or port speed and results in expected traffic loss. The major objective is to ensure critical priority traffic is uninterrupted even during periods of congestion.

During key congestion events, we observed the following:

• Strict-high queue was serviced ahead of low priority queues (up to queue shaped rate). Even during periods of high congestion, when low priority queues are dropping packets, critical flows were guaranteed without any packet loss.
• Low priority queues are guaranteed up to configured transmit-rate (CIR) (strict-high queue is shaped).
• Low priority queues are serviced as WFQ when operating in excess regions and bandwidth is available.
• Low priority remainder queue was granted a transmit rate consistent with leftover bandwidth.
• Scheduler percentages correctly inherit the configured port-shaper as port speed.
• Queue shaping rate is deducted from total bandwidth with transmit-rates applied to the remaining bandwidth.
• Priority hierarchies are honored across and within VPN services that share common links.

For the full test report with details on all test cases, contact your Juniper Networks representative.

Latency Budgets

5G xHaul infrastructure defines strict latency budgets and particularly in the Fronthaul segment where supporting ultra-low latency flows are required. Total budget factors elements such as fiber length, connected devices, and transport design. O-RAN mandates a maximum of 100µs Fronthaul one-way latency from O-RU to O-DU, with each device ~≤10µs. But operations are demanding device latency closer to ~5-6µs. This is a massive paradigm shift from the requirements of earlier 4G architectures.

First, we looked at how ACX7024 performs, taking latency measurements as a standalone platform. Then, we validated how the complete Fronthaul and MBH infrastructure performs with ACX7024 as the CSR.

Topology 1, shown in Figure 2 , was used to validate the performance of the ACX7024 device as the CSR. It offers the most accurate representation of the ACX7024's individual performance without considering additional hops in the network path. The traffic is generated by Ixia, excluding self-latency. We simulated critical traffic flows that represent eCPRI using burst or continuous streams, with packet sizes of 64b, 512b, and 1500b.

Table 4 displays the latency measurements of the ACX7024. In this scenario, a single-DUT utilizes a bridge-domain, with traffic mapped to the strict-high queue. However, it is worth noting that there is a minimal difference observed whether the queue is set to strict-high or low when there is no congestion.

Table 4: Single DUT Latency Measurements

DUT	Queue Priority	Min (µs) Latency	Ave (µs) Latency	Max (µs) Latency	Frame Size	Traffic Pattern	Port
ACX7024	SH	5.44µs	5.46µs	5.94µs	64b	Continuous	10G
ACX7024	SH	5.33µs	5.37µs	5.84µs	512b	Continuous	10G
ACX7024	SH	4.62µs	4.65µs	6.12µs	1500b	Continuous	10G
ACX7024	SH	5.44µs	5.47µs	6.03µs	64b	Burst	10G
ACX7024	SH	5.34µs	5.37µs	5.68µs	512b	Burst	10G
ACX7024	SH	4.63µs	4.66µs	5.87µs	1500b	Burst	10G

For complete outputs of all latency measurements, contact your Juniper Networks representative.

Topology 2, shown in Figure 3 , was used to measure performance across the xHaul, including both CSR and HSR devices in the Fronthaul segment. The ACX7024 is the CSR DUT and the ACX7509 is the HSR.

The Fronthaul segment consists of three hops:

1. CSR ACX7024

1. HSR ACX7509
2. O-DU QFX5110-48S/O-DU.

EVPN single-homed services are between ACX7024 and ACX7509 with QFX being Layer 2 passthrough.

The Midhaul to Backhaul segment (L2Circuit and L3VPN) consists of six hops:

1. CSR ACX7024 (start)

1. HSRs ACX7100-32C and ACX7509
2. AG2 MX204s
3. AG3 MX480/MX10003
4. Core PTX10001-36MRs
5. SAG with MX304 (end)

Table 5 compares the Fronthaul and Midhaul to Backhaul performance across different service types terminating on the ACX7024 CSR. Total latency factors number of hops, for example EVPN-VPWS with three hops measured 14.63µs, amounting to 4.9µs per hop in the Fronthaul segment.

For the full test report, with complete results detailing the minimum/average/maximum latency across all featured feature types (EVPN, L2VPN, L2Circuit, and L3VPN), contact your Juniper Networks representative.

Table 5: Latency Measurements (No Congestion)

Service Type	Queue Priority	Min (µs) Latency	Ave (µs) Latency	Frame Size	Traffic Pattern	Port	Segment	Hop #
EVPN-VPWS	SH	10.53µs	14.63µs	64b	Continuous	10G	FH	3
EVPN-VPWS	SH	11.46µs	16.26µs	512b	Continuous	10G	FH	3
EVPN-VPWS	SH	11.82µs	18.47µs	1500b	Continuous	10G	FH	3
EVPN-VPWS	SH	10.51µs	13.56µs	64b	Burst	10G	FH	3
EVPN-VPWS	SH	11.45µs	15.19µs	512b	Burst	10G	FH	3
EVPN-VPWS	SH	11.63µs	17.38µs	1500b	Burst	10G	FH	3
EVPN-VPWS	SH	--	11.5µs	512b	Continuous	100G	FH	3
EVPN-VPWS	SH	--	15µs	512b	Continuous	1G shaper	FH	3
L2Circuit	LOW	51.19µs	72.8µs	64b	Continuous	10G	MBH	6
L2Circuit	LOW	50.23µs	77.05µs	512b	Continuous	10G	MBH	6
L2Circuit	LOW	49.21µs	82.35µs	1500b	Continuous	10G	MBH	6
L2Circuit	LOW	50.96µs	70.42µs	64b	Burst	10G	MBH	6
L2Circuit	LOW	50.09µs	74.94µs	512b	Burst	10G	MBH	6
L2Circuit	LOW	49.07µs	80.81µs	1500b	Burst	10G	MBH	6
L2Circuit	LOW	--	65.8µs	512b	Continuous	100G	MBH	6
L2Circuit	LOW	--	122.9µs	512b	Continuous	1G shaper	MBH	6
L3VPN	LOW	38.31µs	100.72µs	64b	Continuous	10G	MBH	6
L3VPN	LOW	40.53µs	106.30µs	512b	Continuous	10G	MBH	6
L3VPN	LOW	40.42µs	132.37µs	1500b	Continuous	10G	MBH	6
L3VPN	LOW	--	99.9µs	512b	Continuous	100G	MBH	6
L3VPN	LOW	--	146.5µs	512b	Continuous	1G shaper	MBH	6

The priority queue delivers strict latency performance compared to low priority queues.