IBM i

IBM i 

A space for professionals working with IBM’s integrated OS for Power systems to exchange ideas, ask questions, and share expertise on topics like RPG and COBOL development, application modernization, open source integration, system administration, and business continuity.


#Power


#IBMi
#Power
#Operatingsystems
#Servers
 View Only

Upgrading to IBM i 7.5 in the Cloud: When Your Gateway Starts Disappearing

By Javier Herrera Cuellar posted 09/29/26 12:50 PM

  

Lessons learned from a production upgrade and a one-line fix recommended by IBM Support

The context

Recently I supported a client running IBM i on IBM Power in the cloud. The environment has two partitions: production and contingency, kept in sync with real-time replication. We upgraded the production partition from IBM i 7.4 to 7.5, while the contingency partition stayed on 7.4 for the first phase.

The upgrade itself went as planned. A few days later, however, users started to report intermittent disconnections. Sessions dropped, and for several minutes nothing could reach the system through the network gateway. Then, just as suddenly, everything worked again.

The symptom

The key evidence was in the history log. Every disconnection matched a message like this:

TCP2615 - TCP/IP gateway x.x.x.x not available.
Cause: The TCP/IP gateway is no longer reachable from this system.
Connections routed through the gateway may fail if another route
to the connection destinations cannot be found.

Intermittent network problems are frustrating because everything looks healthy when you check. The history log told a different story: the system itself was deciding, again and again, that the gateway was dead.

What is Dead Gateway Detection?

IBM i includes a mechanism called Dead Gateway Detection. Its purpose is good: when the next-hop gateway of a route stops responding, the system marks it as unavailable and reroutes traffic through an alternate gateway, if one exists.

In practice, when TCP has retransmitted the same data several times, the system probes the gateway with PING or ARP requests. If the gateway does not answer in time, every route that uses it is marked as unavailable and the connections through it are closed. The system then keeps probing the "dead" gateway periodically, and only when it answers again are the routes restored.

Why it was hurting us

The client's partition had a single default route (*DFTROUTE) pointing to one gateway. There was no alternate path to fail over to.

That meant Dead Gateway Detection could not protect anything. A network hiccup of a few seconds was enough for the system to declare the gateway dead, drop every connection, and wait for the next successful probe. A short blip in the network turned into an outage of several minutes.

The fix

IBM Support pointed us to their official recommendation: in a single-gateway environment, disable Dead Gateway Detection. Keep it enabled only when you have redundant routes through different gateways.

Gateway environment Route configuration IBM recommendation
Single gateway Same next hop CHGTCPA IPDEADGATE(*NO)
Multiple gateways Different next hops CHGTCPA IPDEADGATE(*YES) (default)

Before changing anything, we confirmed the route configuration with CFGTCP, option 2, and checked that production had only one default route. Then we applied the change through the client's change control:

CHGTCPA IPDEADGATE(*NO)

Measure before and after with SQL

To prove that the change worked, you need data, not impressions. This query lists every gateway event in the history log for the last seven days:

SELECT MESSAGE_TIMESTAMP, MESSAGE_ID, MESSAGE_TEXT
  FROM TABLE(QSYS2.HISTORY_LOG_INFO(
         START_TIME => CURRENT TIMESTAMP - 7 DAYS)) H
 WHERE MESSAGE_ID LIKE 'TCP26%'
 ORDER BY MESSAGE_TIMESTAMP;

Run it before the change to build a baseline, and again afterwards to compare. It gives you exact timestamps you can correlate with user complaints, network logs, or the cloud provider's events. This query, and others for performance baselines and PTF checks, are available in my open-source ibmi-sql-toolkit repository (https://github.com/javierh77/ibmi-sql-toolkit).

The result

After disabling Dead Gateway Detection, the partition ran for more than a week without a single gateway disconnection, and IBM Support closed the case.

As IBM Support clearly explained, this does not fix the underlying network issue. If the network has short interruptions, they still happen. The difference is that an interruption of a few seconds no longer becomes an outage of several minutes.

Lessons learned

  1. Read the history log first. Intermittent problems leave traces. Message TCP2615 turned a vague complaint into a precise diagnosis.
  2. Know your route topology. Dead Gateway Detection makes sense only with redundant gateways. With a single default route, it can amplify short network glitches into long outages.
  3. Treat the fix as mitigation, not as root cause. Keep investigating the network layer. In virtualized environments, IBM Support suggests checking the Virtual I/O Server and the shared Ethernet adapter, and capturing a trace while the problem occurs.
  4. Measure with SQL. IBM i Services turn the history log into data you can query, compare, and share.

A short checklist for your next IBM i upgrade

  • Review the TCP/IP routes and the IPDEADGATE setting before the upgrade
  • Save a baseline of gateway events from the history log
  • Watch the history log for TCP26xx messages during the first days after the upgrade
  • Validate network behavior with your cloud or network team, not only the IBM i side

Have you seen similar behavior after an upgrade or a move to the cloud? I would like to hear how you handled it in the comments.

Reference: IBM i Dead Gateway Detection - Overview & Recommendations

0 comments
8 views

Permalink