ECMP Path Capacity Calculator
Estimate ECMP usable capacity from healthy paths, link rate, demand, hash efficiency, headroom, failure runway, and largest-flow risk.| Metric | Value | Detail | Copy |
|---|---|---|---|
| {{ row.metric }} | {{ row.value }} | {{ row.detail }} |
| Check | Signal | Recommendation | Copy |
|---|---|---|---|
| {{ row.check }} | {{ row.signal }} | {{ row.recommendation }} |
| Additional failures | Healthy paths | Effective capacity | Spare after demand | Status | Copy |
|---|---|---|---|---|---|
| {{ row.failures }} | {{ row.healthyPaths }} | {{ row.effectiveCapacity }} | {{ row.spareAfterDemand }} | {{ row.status }} |
Introduction:
Redundant network paths do not automatically behave like one large pipe. A fabric may advertise several equal-cost next hops for the same destination, yet traffic still reaches those paths through a forwarding choice made packet by packet or flow by flow. Capacity planning for equal-cost multipath, or ECMP, has to account for the paths that are actually healthy, the speed of each member, the traffic demand, and how evenly the hash can spread flows across the group.
ECMP is most useful when many independent conversations are crossing the same route. Web requests, tenant traffic, storage replication sets, overlay tunnels, and service-to-service calls can spread well when the packet headers provide enough variety for the hash. A few very large transfers can behave differently. If one backup stream or tunnel lands on one member path, the rest of the group may still have spare capacity while that one member is overloaded.
- Configured paths
- The equal-cost next hops the design expects to use when the route is fully available.
- Healthy paths
- The members currently usable after failures, drains, withdrawals, and policy changes.
- Hash efficiency
- A planning discount for uneven flow placement, low entropy, polarization, or uncertain telemetry.
- Elephant flow
- A single conversation, tunnel, backup, or replication stream large enough to threaten one member path.
The practical mistake is to multiply the design path count by link speed and stop there. That number ignores maintenance drains, route withdrawals, failed optics, platform path limits, and the reserve needed for bursts or measurement error. It also assumes the flow mix is friendly enough for the hash to use the group evenly.
An ECMP capacity estimate is a planning aid for change windows, tenant turn-ups, route migrations, and failure reviews. It can show that a path group is short before a maintenance drain starts, or that demand fits only if hash spread stays favorable. It cannot prove that queues, microbursts, asymmetric traffic, convergence behavior, or device-specific hashing will be healthy under live load.
How to Use This Tool:
Start with the active route state, then rerun the same case for the failure or maintenance state you need to approve.
- Set Total ECMP paths to the configured equal-cost next-hop count for the route or prefix.
- Set Healthy paths to the members currently usable in forwarding. Use live route or forwarding state when links are failed, drained, or withdrawn.
- Enter Per-path capacity in Gbps and Aggregate demand for the load that must fit across the active group.
- Enter Largest single flow when one tunnel, backup job, replication stream, or tenant conversation might be larger than one member path. Use 0 only when single-flow risk is outside the check.
- Set Hash efficiency from measured flow distribution when telemetry exists. Use a lower percentage when entropy is uncertain or a small number of large flows dominates demand.
- Set Target headroom for burst, measurement error, and route churn. In Advanced, add Planned path removals before a maintenance drain.
- Confirm the Capacity Ledger for the numeric fit, the Hash Risk Brief for warnings, the Failure Runway for extra path-loss tolerance, and the two charts for a visual check of capacity against demand.
Interpreting Results:
Effective ECMP capacity is the planning value to compare with demand. It starts from healthy wire capacity and applies the selected hash-efficiency discount. A group can have plenty of configured wire capacity while effective capacity is thin because paths are missing or flow placement is uneven.
Spare after demand and Reserve against target answer separate questions. Positive spare means the modeled demand fits. Negative reserve against target means demand fits only by consuming the selected safety margin.
| Cue | Meaning | Practical response |
|---|---|---|
| target met | Effective capacity is greater than or equal to demand plus the selected headroom. | Check that demand, path health, and hash efficiency came from a comparable traffic window. |
| reserve thin | Effective capacity covers raw demand but not the headroom target. | Delay drains or traffic growth unless more capacity, lower demand, or better evidence is available. |
| capacity short | Effective capacity is below aggregate demand. | Do not rely on the path set until capacity, demand, path health, or hash assumptions change. |
| single-flow tight | The largest modeled flow is above 85% of one member path. | Inspect flow placement and decide whether the traffic can be sharded, moved, or rate-limited. |
| single-flow over path | The largest modeled flow exceeds one member path. | Ordinary ECMP hashing will not split that one flow across every member. |
Failure Runway shows modeled capacity after each additional path failure. Rows marked Target covered still meet the reserve target, Demand only carry raw demand without the selected reserve, and Short fall below raw demand. The ECMP Capacity Stack chart compares configured wire, healthy wire, effective capacity, demand, and demand plus headroom. The Path Failure Curve shows how quickly the group loses margin as members disappear.
A favorable result still needs operational evidence. Vendor hash fields, tunnel entropy labels, adaptive load balancing, queue policy, link oversubscription, return-path asymmetry, and microbursts can all change live behavior. Treat the result as a capacity approval check, then compare it with interface counters, flow telemetry, routing state, and platform limits.
Technical Details:
ECMP separates path eligibility from traffic placement. Routing decides which equal-cost next hops can be installed for a destination. Forwarding then chooses a member for each packet or flow according to the platform's load-balancing method. Flow-aware hashing is common because it keeps related packets ordered, but that same behavior makes capacity depend on the number and size of flows.
The capacity model treats healthy path count as the active wire ceiling and hash efficiency as a discount against that ceiling. Headroom is applied to demand, not to link rate, because reserve is a requirement above the traffic that must be carried. Failure runway repeats the same capacity equation after additional members are removed.
Formula Core:
The main equations convert path count, member speed, and hash efficiency into an effective Gbps value, then compare that value with raw demand and demand plus reserve.
| Symbol | Meaning | Unit or source |
|---|---|---|
T |
Configured ECMP path count. | Whole paths, at least 1. |
H |
Healthy ECMP path count. | Whole paths from 0 through T. |
C |
Usable capacity of one member path. | Gbps, kept above zero. |
η |
Hash efficiency allowance. | Percent from 1 to 100. |
D |
Aggregate demand. | Gbps, at least 0. |
R |
Target headroom above demand. | Percent from 0 to 200. |
f |
Additional path failures or removals in the runway row. | Whole paths from 0 through H. |
With 7 healthy paths at 10 Gbps each, healthy wire capacity is 70 Gbps. At 82% hash efficiency, effective capacity is 57.40 Gbps. If aggregate demand is 55 Gbps and headroom is 15%, target demand is 63.25 Gbps, so the group carries raw demand but misses the reserve target by 5.85 Gbps.
| Input | Supported boundary | Effect on the model |
|---|---|---|
| Total ECMP paths | Rounded down and kept at one or more. | Sets configured wire capacity and caps healthy paths. |
| Healthy paths | Rounded down and kept from zero through total paths. | Sets active wire capacity before hash efficiency is applied. |
| Per-path capacity | Kept above zero. | Multiplies every configured, healthy, maintenance, and failure-runway path count. |
| Hash efficiency | Kept from 1% through 100%. | Discounts healthy wire capacity for uneven flow placement. |
| Target headroom | Kept from 0% through 200%. | Raises the demand threshold that effective capacity must cover. |
| Planned path removals | Rounded down and capped at the healthy path count. | Shows maintenance capacity after the planned drain. |
The largest-flow check is deliberately separate from aggregate capacity. Four 10 Gbps members at 90% hash efficiency produce 36 Gbps of effective capacity, which can be enough for 24 Gbps of total demand. A single 12 Gbps flow still exceeds one 10 Gbps member, so a normal flow hash cannot make that one conversation use all four paths.
Boundary comparisons are inclusive. Effective capacity greater than or equal to demand plus headroom is target met. Effective capacity greater than or equal to raw demand but below the target is reserve thin. Effective capacity below raw demand is capacity short. Single-flow status changes to single-flow tight above 85% of one member path and single-flow over path when the largest flow is greater than the member path capacity.
Accuracy and Privacy Notes:
The calculation runs in the browser from the numbers entered on the page. It does not need a live router query, topology discovery, packet capture, or server-side lookup to produce the capacity ledger, risk brief, failure runway, charts, and JSON output.
Use measured telemetry for operational decisions. Interface counters, sampled flows, route state, optics, platform ECMP limits, tunnel entropy, queue policy, and maintenance procedures can all change real margin. Low hash efficiency, missing paths, or single-flow warnings should trigger network inspection rather than automatic diagnosis.
Worked Examples:
Seven of eight paths are healthy:
With 8 total paths, 7 healthy paths, 10 Gbps per path, 55 Gbps demand, 82% hash efficiency, and 15% headroom, healthy wire capacity is 70.00 Gbps and effective capacity is 57.40 Gbps. Demand fits with 2.40 Gbps spare, but the 63.25 Gbps reserve target is not met, so the result is reserve thin.
All paths restored with better spread:
Changing the same case to 8 healthy paths and 90% hash efficiency raises effective capacity to 72.00 Gbps. The reserve target is covered. One additional path failure leaves 63.00 Gbps, which still covers the raw 55.00 Gbps demand but falls just below the 63.25 Gbps headroom target.
Maintenance drain should be delayed:
A group with 10 healthy paths at 25 Gbps each, 180 Gbps demand, 85% hash efficiency, and 20% headroom has 212.50 Gbps effective capacity. That already misses the 216.00 Gbps reserve target. Removing two paths for maintenance drops effective capacity to 170.00 Gbps, which is short of raw demand.
Aggregate fit hides a single-flow problem:
Four 10 Gbps members at 90% hash efficiency provide 36.00 Gbps effective capacity. A 24.00 Gbps aggregate load with 10% headroom fits, but a 12.00 Gbps largest single flow is still larger than one 10 Gbps member. The aggregate model passes while the single-flow warning remains critical.
FAQ:
Why not multiply every configured path by link speed?
Configured paths are the design maximum, not necessarily the active forwarding set. Failed, drained, or withdrawn paths do not carry traffic, and hash imbalance can make healthy wire capacity less useful than the raw total.
What should I use for hash efficiency?
Use measured distribution when flow telemetry exists. Without measurements, 70% to 90% is a cautious planning range for uncertain entropy. Values near 100% fit many-flow traffic known to balance well.
Can a single large transfer use every ECMP member?
Usually no. Flow-aware ECMP keeps packets for one flow on one member to avoid reordering. A large workload can use more of the group only if it opens multiple flows that hash to different members or uses another load-splitting design.
Why did an entered value change?
Path counts are whole-path quantities, percentages have supported ranges, and negative demand or flow values are not useful for this model. The input audit names any clamping or rounding applied to the calculation.
Does target met mean maintenance is safe?
It means the entered numbers meet the selected capacity model. Maintenance still needs change-control checks for route convergence, device limits, optics, queues, rollback time, and traffic peaks during the window.
Glossary:
- ECMP
- Equal-cost multipath forwarding, where multiple next hops with equal routing cost can be used for one destination.
- Flow hash
- A forwarding selection method that maps packet-header fields to one member path so related packets stay together.
- Healthy wire capacity
- Healthy path count multiplied by per-path capacity before the hash-efficiency discount.
- Effective ECMP capacity
- Healthy wire capacity after the selected hash-efficiency allowance is applied.
- Headroom
- Reserve capacity above demand for bursts, measurement error, failures, or maintenance uncertainty.
- Failure runway
- The modeled ability to keep covering demand or demand plus headroom after additional paths fail.
References:
- RFC 2991: Multipath Issues in Unicast and Multicast Next-Hop Selection, RFC Editor, November 2000.
- RFC 2992: Analysis of an Equal-Cost Multi-Path Algorithm, IETF Datatracker, November 2000.
- Understanding the algorithm used to load balance traffic on MX Series routers, Juniper Networks Junos OS Documentation.