AEON on a bank-shaped estate
A 1,000-rack primary data centre, a 200-rack disaster-recovery site, 1,008 branches in 12 zones, 1,554 ATMs and 99 ultra-small
banking points: 10,976 devices on one machine with 4 GB of memory. Every figure below was produced by AEON through its own interface.
Devices10,976in 2,335 locations
Primary data centre1,000 racksplus a 200-rack recovery site
Branches1,008in 12 zones fed by zone hubs
ATMs1,554each monitored individually
Samples ingested552,696two hours of history in 1.5 s
Diagnosis cycle1.7–8.9 s136 MB peak memory
Alerts for a zone outage1287 devices went quiet; the hub alone was reported
Building the estate
| Locations created | 2,335 | 137.8 s |
|---|
| Devices imported | 10,976 | 12.4 s |
|---|
| Relationships written | 18,317 | 0.9 s |
|---|
| Readings ingested | 552,696 | 1.5 s |
|---|
What an operator waits for
| Primary data centre, 1,000 racks | 0.43 s |
|---|
| Disaster-recovery site, 200 racks | 0.21 s |
|---|
| One data hall, 100 racks | 0.38 s |
|---|
| One rack elevation | 0.22 s |
|---|
| Whole estate | 0.47 s |
|---|
| One zone of 112 branches | 0.42 s |
|---|
| One branch | 0.23 s |
|---|
| Estate-wide health rollup | 1.01 s |
|---|
| Firing alerts, one page | 0.04 s |
|---|
| Incident grouping, whole estate | 0.1 s |
|---|
| Placement across 1,200 racks | 0.3 s |
|---|
| Platform self-check | 0.25 s |
|---|
Faults we injected, and what AEON did
| A zone hub fails | One alert, naming the hub. The 286 branch routers, switches and ATMs behind it were held back, kept and visible, with the cause stated. The 28 branches with a second connection stayed reachable. |
|---|
| A cooling unit degrades | Named as the root cause behind 250 servers, as one incident. Intake temperature predicted to reach 35 °C in about 18 minutes. |
|---|
| Internet banking slows | Alerted at 4.8 seconds per response against a 2-second rule. |
|---|
| Replication to the recovery site slows | Alerted at 2.9 seconds against a 2-second rule. |
|---|
| ATMs on poor links drop in and out | Not alerted. A device must be down for a sustained period before AEON calls it unreachable, and there is no rule yet for a link that keeps flapping. We are building one. |
|---|
Why the first row matters. When a zone hub fails, every branch behind it goes silent at once. A monitoring
system that reports each one separately wakes the on-call engineer with nearly three hundred alerts for a single fault.
AEON reported the hub and held the rest back with the reason attached.
What the run found in AEON itself. Alert suppression did not yet follow the network, so that hub failure would have
produced an alert per branch; an alert rule could accept a filter it did not support and quietly ignore it; and a size limit split one
cooling failure into two incidents. All three were fixed and tested before these figures were taken.
Real and simulated. The platform and its databases were real and running. The estate and its readings were generated:
no physical equipment was involved and nothing was controlled. The event store was not running for this test, and AEON's own self-check
reported that correctly.