NeuBird
LoginDemo

NeuBird for Cisco SD-WAN

Cisco SD-WAN troubleshooting, from “the app is slow” to the circuit.

NeuBird is the AI SRE. It reads Cisco SD-WAN Manager alongside your application, deploy and cloud telemetry, and answers the question every branch slowdown starts with: the app or the WAN path? NeuBird names the site, the circuit and the change, with root cause in under 5 minutes at 94% accuracy.

The question

Is it the app or the WAN path?Four answers, from one investigation.

When users at a branch say the app is slow, the application team sees healthy services and the network team sees a fabric that is mostly green. The answer sits between them, in the BFD sessions, app-route statistics and alarms in SD-WAN Manager, read against what the application saw at the same moment.

  • App or WAN path?

    Application latency at one site, read against tunnel health at that site, separates a slow service from a degraded path.

  • Which site?

    Every other site and the cloud stay flat, so the problem narrows to one branch and one WAN Edge.

  • Which circuit?

    BFD sessions and app-route statistics per color show which transport is losing packets: mpls, biz-internet or lte.

  • Which change?

    Recent deploys, config group pushes and ISP faults are lined up on one timeline, and the innocent ones are ruled out.

Connect

What does NeuBird read from Cisco SD-WAN Manager?The whole fabric, through GET calls only.

NeuBird connects to Cisco SD-WAN Manager, formerly vManage, with one dedicated user in a group that can read every feature and write none. It reads devices, alarms, BFD sessions, OMP and BGP state, control connections and application-route statistics, and queries them in place. Every call is an HTTP GET, so the fabric stays exactly as your team configured it, and NeuBird deploys in under one hour.

  1. 1

    A dedicated user

    One SD-WAN Manager user in a group with Read and no Write, such as the built-in operator group.

  2. 2

    The whole fabric in view

    Devices, alarms, BFD sessions, OMP and BGP state, control connections and app-route statistics.

  3. 3

    GET calls only

    Every call is an HTTP GET. The fabric stays exactly as your team configured it.

  4. 4

    Synced

    NeuBird queries SD-WAN Manager in place, with no bulk telemetry stored.

Connect Cisco SD-WAN ManagerNot connectedSynced
Connection
SD-WAN Manager URLhttps://sdwan-manager.example.com
Usernameneubird-ops
User groupoperator · Read on every feature, no Write
Password••••••••••••
Write calls · noneFabric changes stay with your network team, staged by NeuBird behind an approval gate.
Synced · 1 SD-WAN Manager · 64 WAN Edges · 3 Controllers · 2 Validators
What NeuBird reads
GETDevicesWAN Edges, Controllers, Validators · site, system IP, version, reachability
GETAlarms and eventsBFD_TLOC_Down, Control_Node_Down, OMP_Node_Down and the rest
GETBFD sessionsPer tunnel: local and remote color, state, transitions, uptime
GETOMP and BGPPeers, routes and TLOCs advertised across the overlay
GETControl connectionsEach edge to its SD-WAN Controllers and Validator
GETApp-route statisticsLoss, latency and jitter per tunnel and color, against SLA classes

Resolve · when a site degrades

How does NeuBird find the circuit behind a slow app?It reads the fabric the way your network team would.

In this sample incident, the biz-internet circuit at branch DAL-BR-014 starts dropping packets at 9:02. SD-WAN Manager records it at once: the BFD sessions on that TLOC flap, its app-route statistics break the business-critical SLA class, and Application-Aware Routing starts moving traffic to mpls. The other four sites stay green, which is why the fabric looks healthy at a glance.

  1. 1

    The overlay

    Five Dallas branches, each with mpls and biz-internet tunnels to two hub sites.

  2. 2

    One site degrades

    At DAL-BR-014, every biz-internet tunnel turns amber, then starts to flap.

  3. 3

    BFD tells the story

    Three biz-internet sessions on edge-dal-01 go down, 14 transitions since 9:02. mpls holds.

  4. 4

    The alarm, in context

    BFD_TLOC_Down on one TLOC at one site, not a fabric-wide outage.

CiscoSD-WAN Manager · Dallas regionSample fabric
DAL-DC-01edge-dc-01 · edge-dc-02AUS-DC-01edge-aus-01DAL-BR-011DAL-BR-012DAL-BR-014DAL-BR-016DAL-BR-019
mplsbiz-internet5 of 5 sites healthyDAL-BR-014 · partial · 3 of 6 BFD sessions down
Ciscoedge-dal-01 · BFD sessionsSystem IP 10.255.14.1
PeerLocal / remote colorStateTransitions
edge-dc-0110.255.0.1biz-internet→ biz-internetupdown214
edge-dc-0110.255.0.1mpls→ mplsup2
edge-dc-0210.255.0.2biz-internet→ biz-internetupdown214
edge-dc-0210.255.0.2mpls→ mplsup2
edge-aus-0110.255.1.1biz-internet→ biz-internetupdown214
edge-aus-0110.255.1.1mpls→ mplsup2
BFD_TLOC_Down · edge-dal-01 · TLOC biz-internet · 14 since 9:02 · Major

Resolve · the investigation

App side and WAN side, in one investigation.Root cause in under 5 minutes, every link cited.

NeuBird queries 15+ sources in parallel. It starts from the symptom in Datadog, rules out the app, the cloud and the 8:55 deploy, then reads alarms, BFD sessions and app-route statistics in SD-WAN Manager. The fix needs a fabric change, so NeuBird stages it for the network team behind a configurable human approval gate. The conclusion is saved to memory, with no bulk telemetry stored, so the next fault on that circuit is recognized on sight.

  1. 1

    Start from the symptom

    A Datadog monitor fires: checkout-web p95 is 2.4 s, from one branch.

  2. 2

    App side first

    Other sites are flat, the cloud is flat, and the 8:55 deploy is ruled out.

  3. 3

    Then the WAN path

    Alarms, BFD sessions and app-route statistics in SD-WAN Manager point at one circuit.

  4. 4

    Cited root cause

    The causal chain, the evidence for each link, and a change staged for approval.

Investigation · checkout-web latency, DAL-BR-014Querying in parallelRoot cause in 4m 08s
#SourceQuestionAnswer
1Datadog APMcheckout-web p95 by site · 8:30 to 9:14Running…DAL-BR-014: 2.4 s · every other site: 310 ms
2GitHubDeploys to checkout-api · last 2 hoursRunning…v3.18 at 8:55 · ruled out, other sites flat
3AWSALB target response time, checkout-apiRunning…Flat at 42 ms
4SD-WAN ManagerAlarms · site DAL-BR-014Running…14 × BFD_TLOC_Down, biz-internet, since 9:02
5SD-WAN ManagerBFD sessions · edge-dal-01Running…biz-internet 3 of 3 flapping · mpls 3 of 3 up
6SD-WAN ManagerApp-route statistics · edge-dal-01 · 9:00 to 9:15Running…biz-internet: loss 6.8%, latency 182 ms, jitter 41 ms
7SD-WAN ManagerControl connections and OMP · edge-dal-01Running…Controllers up, OMP routes stable
Root causeConfidence: high · 4 sources agree

WAN path, not the app. Packet loss on the biz-internet circuit at DAL-BR-014 makes the BFD sessions on edge-dal-01 flap, and Application-Aware Routing keeps moving checkout traffic between colors.

ISP loss on biz-internet, 9:02BFD flaps, 14 transitionsApp-Aware Routing reroutes, again and againcheckout-web p95 2.4 s at DAL-BR-014
Evidence: alarms, BFD sessions and app-route statistics in SD-WAN Manager · Datadog APM · GitHub · AWS
Staged for the network teamPolicy: Recommend
  1. 1Prefer mpls for business-critical traffic at DAL-BR-014Config group change · applied by the network team in SD-WAN Manager
  2. 2Send the ISP the loss evidence for circuit DAL-BR-014-INETTicket drafted with BFD and app-route statistics attached
Auditable approval gate · awaiting Andrew LeeApproved by Andrew Lee · 9:21 AM
Change applied by the network team. checkout-web p95 at DAL-BR-014 back to 320 ms at 9:34.Saved to memory: conclusion, cited evidence and approval. No bulk telemetry stored.

Prevent · before the page

Can NeuBird catch a degrading circuit before users notice?Drift flagged 30 to 60 minutes before the SLA breaks.

Application-Aware Routing steers traffic once a tunnel breaks its SLA class. NeuBird reads the same app-route statistics for the trend: loss and jitter on a TLOC climbing steadily while every number is still inside the SLA. It flags the tunnel 30 to 60 minutes before the breach, matches the pattern against faults it has already resolved, and stages the move while users are still unaffected.

  1. 1

    Every tunnel, every color

    App-route statistics for edge-hou-01, against the business-critical SLA class.

  2. 2

    mpls is clean

    Loss near zero, latency 52 ms, jitter 3 ms, all morning.

  3. 3

    biz-internet drifts

    Loss and jitter climb steadily, still inside the SLA, so nothing alerts yet.

  4. 4

    Flagged 45 minutes early

    NeuBird projects the breach at 11:05 and stages the move before users feel it.

CiscoApp-route statistics · edge-hou-01 · HOU-BR-007SLA class business-critical
business-criticalLoss ≤ 1%Latency ≤ 150 msJitter ≤ 30 ms
Loss (%)
Latency (ms)
Jitter (ms)
mplsbiz-internetSLA thresholdProjection
NeuBird · PreventFlagged 10:20 · 45 minutes ahead

biz-internet on edge-hou-01 is on course to breach business-critical at about 11:05: jitter 23.2 ms rising toward 30 ms, loss 0.6% rising toward 1%. mplsis clean. The signature matches last week's ISP fault at DAL-BR-014, from memory.

Recommended: prefer mpls for business-critical traffic at HOU-BR-007 now, and send the ISP the trend.Auditable approval gate · awaiting Andrew Lee

Operate · between incidents

Fabric hygiene, every week.The work that never makes it onto a ticket.

Between incidents, NeuBird reads device inventory, alarms and control connections across the fabric and posts a weekly review to your network channel: alarm rules that only make noise, sites with a single transport, WAN Edge certificates near expiry, software versions drifting apart and edges short of a Controller connection. Each finding comes with the recommended next step, and the Suggest, Recommend and Act policy dial is set per action.

  1. 1

    Alarm noise

    Alarms that fire every week and never lead to an incident, with the rule to tune.

  2. 2

    Single points of failure

    Sites running one transport, with no backup TLOC to fail over to.

  3. 3

    Certificates and versions

    WAN Edge certificates near expiry and software versions that drift apart.

  4. 4

    Control plane gaps

    Edges holding fewer Controller connections than they should.

Fabric hygiene · weekly review64 WAN Edges · 5 findings
AreaFindingRecommended
Alarm noiseInterface_State_Change on the lte color at 3 WAN Edges: 412 alarms in 7 days, none linked to an incidentTune the alarm rule
No backup transport4 sites run a single TLOC, mpls only: SAT-BR-003, ELP-BR-002, LBB-BR-001, MAF-BR-004Add a second transport
Certificates2 WAN Edge certificates expire within 30 days: edge-okc-01, edge-tul-02Renew before expiry
Version drift11 of 64 WAN Edges run 17.9.4a; the other 53 run 17.12.3aSchedule the upgrade
Control planeedge-elp-01 has held 1 of 2 Controller connections since TuesdayCheck the second path
Read from SD-WAN Manager device inventory, alarms and control connections · posted to #netops every Monday

Across the network

Which other network tools does NeuBird work with?The same investigation, wherever the path runs.

NeuBird queries 50+ integrations in place, so the SD-WAN evidence sits beside your security edge, firewalls and campus network in one investigation. See every solution.

FAQ

Common questions

What does NeuBird do for Cisco SD-WAN?

NeuBird is the AI SRE. When a branch reports that an application is slow, NeuBird reads Cisco SD-WAN Manager alongside your application, deploy and cloud telemetry and answers whether the cause is the app or the WAN path, at which site, on which circuit and after which change. It isolates root cause in under 5 minutes at 94% accuracy, then keeps going: it flags tunnels drifting toward an SLA breach 30 to 60 minutes early and runs fabric hygiene between incidents.

How does NeuBird connect to Cisco SD-WAN Manager?

With the SD-WAN Manager URL and a dedicated user. In SD-WAN Manager, create a user in a group with Read and no Write on every feature, such as the built-in operator group, and enter its username and password in NeuBird. NeuBird reaches SD-WAN Manager over HTTPS, and the connection is ready when it shows Synced.

Which Cisco SD-WAN data does NeuBird read?

Device inventory, alarms, BFD sessions, OMP and BGP state, control connections to the SD-WAN Controllers and Validators, and application-route statistics: loss, latency and jitter for every tunnel and transport color. NeuBird queries them in place, so nothing is exported from SD-WAN Manager first.

Can NeuBird change our SD-WAN fabric?

Not through this connection. Every call NeuBird makes to SD-WAN Manager is an HTTP GET, so the fabric stays exactly as your team configured it. When a fix needs a policy or config group change, NeuBird stages it with the cited evidence behind a configurable human approval gate, and your network team applies it through its own change process. The Suggest, Recommend and Act policy dial is set per action.

How does NeuBird tell an application problem from a WAN problem?

It compares the two side by side. If application latency rises at one site while every other site and the cloud stay flat, and that site shows BFD sessions flapping or app-route loss, latency or jitter above its SLA class, the cause is the WAN path. If latency rises everywhere after a deploy while the tunnels stay within SLA, the cause is the application.

Does NeuBird store our SD-WAN telemetry?

No. NeuBird queries SD-WAN Manager where the data lives, with no bulk telemetry stored. Its memory holds conclusions, cited evidence and approvals, never raw statistics, logs or traces, so the next time the same circuit misbehaves NeuBird recognizes the pattern.

Does NeuBird work with vManage, vSmart and vBond?

Yes. Cisco renamed vManage to SD-WAN Manager, vSmart to SD-WAN Controller and vBond to SD-WAN Validator. NeuBird connects to SD-WAN Manager, which is the same management plane under its current name, and reads Controller and Validator state through it.

NeuBird for Cisco SD-WAN

App or WAN path, answered in under 5 minutes.

See NeuBird, the AI SRE, read Cisco SD-WAN Manager beside your application telemetry and name the site, the circuit and the change.