Known Limitations

General Routing

• In some corner cases traffic is not scheduled equally between strict priority queues. This can happen in the following scenario: Priority queue configured and is completely utilizing the bandwidth and remaining queues are starved and traffic completely drops on those queues. In this state if we configure a second strict high priority queue traffic is not scheduled equally between strict priority queues. This is hardware specific issue, Q2C specific. PR1577035

• VOQ buffer in Packet Forwarding Engine has been changed from LOW to MID for some protocols. Because of this, frames do not drop as soon as ingress traffic rate crosses the configured pps rate. However, the rate towards CPU is still maintained at configured pps rate although violation counters do not increment. And this is evident from the DdoS Routing Engine-based statistics. PR1592370

• When PTP is enabled, ACX7100-32C port 31 is configured to 1G. A vendor's J2 only supports to 10G, so another vendor's gearbox converts 10G to 1G by rate matching. However, there is one limitation: it can only support frame size of 2000 bytes officially, so jumbo frame cannot be supported on port 31 when PTP mode is enabled. PR1593015

• Degradation is due to both unit programming during synchronous install calls. Therefore, programming happens to both units one after the other. PR1597191

• ACX7509, ACX7100-32C and ACX7100-48L do hardware timestamping. However, the peer device (in this case MX Series) does software timestamping, which takes more time resulting in high delays. This issue is seen in all the cases where peer setup does software time stamping. PR1599777

• 1M MPLS label scale is not possible with symmetric traffic testing (where encapsulation and decapsulation are scaled together). In this case, the encapsulation is limited by FEC hardware resource. 52000 scale is tried and tested with symmetric approach. PR1600545

• The queue statistics are not appropriate in the following scenarios:

  1. When DRAM block is asserted in case of continuous congestion. DRAM block is a mechanism to stop traffic from going to DRAM when it is oversubscribed. This prevents packet stuck in the DRAM forever. (DRAM block feature is enabled by default in the ASIC). For this issue we have provided a CLI to disable the DRAM block. You might not see the issue after disabling it.
  2. Packet deleted in VoQ due to non-arrival of credit (Watchdog timer flushing the VoQ).

  Per the vendor, packet deleted in VoQ due to non-arrival of credit (Watchdog timer flushing the VoQ) is not counted as dropped packet. This can happen in two scenarios:

  • The bandwidth is completely configured on a few queues and the other queues do not get any credit.

  • When priority queue is configured and only priority traffic gets scheduled and the other queues do not get any credit and no shaper on priority queue.

  In scenario b) the statistics on the SHP also fluctuates. We recommend to configure shaper on priority queue to prevent starvation of other queues. PR1601320

• ACX7509 has a mesh fabric and it is not a real fabric. There is a flow control mechanism from fabric to CGM. In case of fabric congestion there is flow control and back pressure to the VOQs and packets are dropped in VOQs. These VOQs cannot be mapped directly to fabric congestion as VOQ drops might occur for different reasons. There is no such specific dedicated registers or counters from BCM to provide values for fabric drops. As of now its accounted as queue drops as fabric is back pressuring ingress. PR1601332

• In case of scaled scenarios with route table scale reaching upto their full limit (close to 1M), a few triggers which reprogram the routing table might result in route entries getting missed in hardware. This happens in a worst-case scenarios of prefix distribution. In real-time scenario this issue must not be seen while the internet route scale is around 800,000 which must hold good. PR1604034

• This is the common issue of Junos OS Evolved platform for 400G DAC. On picd restart, the first pcsE has no Xcvr information, and the default Xcvr type is optics, so pfemand configures the port as 400G optics and turn off AN, later on. When the second pcsE arrives, it carries the correct Xcvr information, and turns on AN again. This might flap one link. In the early stage of picd restart, the underlying link is still up, this updates to picd inevitably. Then the first PcsE event comes, and link is down. After the second PcsE event comes, link is up. PR1604454

• Route installation (/32) routes in Packet Forwarding Engine takes longer time to install in LPM, when LEM table limit is reached in the hardware. This issue is observed, only when LEM scale is lesser and LPM scale is on the higher side. when LEM reports 'Table Full', it takes around 1ms to install 1 route in hardware. Since we have around 2.3M routes to be pushed under this condition, the LPM installation will take longer time. PR1604954

• PTP to PTP noise transfer fails for frequencies 1. 0.03125 HZ 2. 0.123125 HZ. PR1608786

• syncE to PTP and syncE to 1pps noise transfer tests fails for frequencies 1. 0.00781 HZ 2. 0.01563 HZ 3. 0.03125 HZ 4. 0.06156 HZ 5. 0.12313 HZ. PR1608866

• On ACX7100-48L, enabling or disabling of PTP TC or BC causes all the interfaces to flap at the same time. PR1609927

• If more than 400 PTP clients are connecting to a single interface configured for PTPoE, PTP delay statistics might not be collected correctly. PR1611130

• Fabtoken is not used by ACX Series platforms except multiline card where similar mechanism of pfetoken is used instead of fabtoken. PR1614670

• Egress firewall ACL has limited lookup width available, so family-any and family ethernet-switching/ccc/vpls/inet6 filters cannot co-exist. PR1617952

• In a system with different type of interfaces (IRB, ccc, rest of interfaces), classifier might not work in specific scenarios. PR1619441

• Egress firewall ACL rules with count or policer actions might be skipped for BUM traffic and might not hit. This is because of ASIC limitation for BUM traffic and applies to all services like L2, L3, EVPN etc. PR1621609

• The learning rate of ACX7509 is same as ACX7100 when the host routes /128 routes are within the limit of ACX7100 scale (375K) or LEM table size limit. More than 50% degradation is observed in Fib learning rate on IPv6 host routes (/128) as compared to ACX7100, only when the scale exceeds 375K routes . PR1624365

• The error message "ACX7509: HWD_FIRMWARE_VERSION_READ_ERROR_NOTICE: Failed to get firmware version for fru: Feb, slot: 1, device: Dpll - Error message " is to be seen, when the feb is offlined and the dpll version is tried to be read. From Cli the FEB slot1 DPLL version is properly read later. FEB 1 DPLL 9 600 6022 OK. PR1625324

• CCM transmission keeps happening (for the UP MEP aggregated Ethernet case) during FPC restart. This is because, aggregated Ethernet interface might not be deleted during the FPC restart. Therefore, the session information that is already programmed in the hardware will not get deleted. As a result, during the FPC restart case, the CCM transmission keeps happening. Due to this reason, sessions are up on the peer. Whereas in case of physical interface, the sessions will get deleted and added back during FPC restart. PR1627369

• Ingress replication is default for multicast. Packet Forwarding Engine can replicate upto 2.4T which is the chip capacity. Inter-Packet Forwarding Engine replication is limited to 2.1T due to fabric bandwidth limitations. Therefore, with ingress replication, it is not possible to achieve total egress traffic across ports more than 2.1T. PR1629280

• 2M is not achievable without errors, 1.5M routes are the appropriate value for the scaled scenarios. PR1630088

• The fields in output of the CLI show pfe statistics traffic, which are not supported in ACX7509, cannot be hidden, as this is a design limitation. The fields continue to show zero value and can be ignored. There is no impact on any feature. PR1630562

• Maximum support logical interface is 16000. Each local XC needs 2 logical interfaces. Therefore, achieved scale will be 8000. PR1636947

• Maximum supported logical interface is 16000. Each pseudowire require 1 logical interface. Therefore, 16000 pseudowire scale will be achieved. PR1637066

• Configd Object-info anomalies are seen after you commit the show platform object-info anomalies app configd configuration statement. PR1635668

Routing Protocols

• IPv6 link-local based BFD sessions work in a centralised BFD mode in ACX7509. So we can not have aggressive BFD timers supported for IPv6 link-local based BFD session. PR1599658