If you already run points of presence elsewhere in Southeast Asia or APAC, you know the pattern: a cache node goes into a new market, and the win-or-lose factor is never the server — it’s the peering underneath it. Get that right, and the cache serves local users on a single hop at a fraction of transit cost. Get it wrong, and a Jakarta box quietly performs worse than the offshore PoP it was meant to replace.
This is a working playbook for regional CDN operators who’ve already decided Indonesia is next and now need the operational detail: how Edge Peering Internet Exchange (EPIX) fits into the plan, what to prepare before the cross-connect is ordered, and when it’s time to move beyond the exchange into bilateral peering. It assumes you already know BGP and exchange peering, the focus here is what’s specific to Indonesia.
One practical advantage worth stating upfront: Digital Edge operates its data centers in Jakarta itself, where EPIX is hosted, so a cache deployed with us reaches the exchange over a short in-building connection rather than a metro-area haul. That distinction matters more than it sounds like once you’re comparing latency and cross-connect lead times across facility options.
Why the Exchange Comes First Here
Operators bring a mature-market instinct with them: negotiate the big bilateral deals first, let the exchange mop up the rest. In Indonesia, that instinct is usually backwards.
The Indonesian eyeball market is concentrated among a handful of large operators, but most of them treat the exchange as their primary peering venue, not a fallback. EPIX’s route servers reach a broad slice of domestic content and access networks through a single port, reach that would otherwise take dozens of separate bilateral negotiations to build. The current participant list, port capacities, and IPv6 coverage are published on EPIX’s PeeringDB record, and it’s worth reading before any commercial conversation starts; the roster changes, and PeeringDB is more current than any vendor deck, including this one.
The sequence that works: join EPIX, accept the route servers, watch traffic for the first 30 days, then add bilateral interconnects only for the networks whose volume earns a dedicated port. Operators who negotiate bilaterals first typically spend months on agreements the exchange would have handled in a week — and the cache underperforms the whole time they’re waiting.
Before You Order the Cross-Connect
Most of what determines whether peering goes smoothly happens before any physical link is built.
PeeringDB is the first thing Indonesian networks check. Fill in your ASN, IPv4/IPv6 addresses at each exchange, prefix counts, traffic levels, policy URL, and real contacts for both peering coordination and a 24×7 NOC. A half-filled record is one of the most common reasons peering requests stall — the other side can’t evaluate what it can’t see.
Make your IRR records match what you’ll actually announce. Indonesian networks filter routinely on Internet Routing Registry data. A prefix that isn’t properly registered gets dropped regardless of what was agreed over email. Check your IRR entries are current before the port goes live, not after.
Resource Public Key Infrastructure (RPKI) matters more than it used to. Several large Indonesian networks now drop RPKI-invalid routes outright. Publish Route Origin Authorisations for every prefix you plan to announce before peering goes live — this is cheap insurance against route leaks and slow convergence, and expensive to fix retroactively.
Publish a peering policy. “Open at exchanges, selective for PNIs above [X] Gbps sustained” is the posture most foreign CDN operators land on, and it’s the one Indonesian counterparties expect. Leaving this ambiguous is what turns a one-week request into a one-quarter negotiation.
Setting Up on EPIX
Pick a port size with headroom. Undersized ports saturate faster than operators expect once cache hits ramp, and re-provisioning mid-ramp is disruptive at exactly the wrong time. EPIX offers 10G, and 100G ports, with 400G available on request. The right starting point for a foreign CDN cache layer is usually 10G with headroom.
Think about where the cross-connect physically lands. A cache inside the same facility as the exchange reaches it over a short internal patch. A cache in a third-party facility inherits longer lead times, one more operational dependency, and extra latency just to reach the exchange. This is one of the more concrete reasons to weigh a Jakarta-based facility choice up front: co-locating in the same building as EPIX turns the cross-connect into a patch panel exercise instead of a metro fiber build.
Default to the route servers. They expose most of the exchange’s participants without per-peer configuration. Filter on prefix length, RPKI validity, and AS-path sanity rather than building a manual peer list. The exception is the small number of networks that don’t participate in the route servers — those still need direct sessions.
Set MED policy deliberately, don’t inherit it. A MED policy tuned for an offshore PoP tends to produce odd path selection in Indonesia, where multiple paths to the same destination exist across different exchanges and transit providers. Set MED values for the Jakarta node explicitly, prefer exchange-learned routes over transit, and check actual path selection in the first two weeks, not after a complaint comes in.
Adding Bilateral Peering — and Knowing When
The exchange covers most of the market; bilaterals exist to handle the volume tails.
The candidates are the networks whose share of your traffic crosses the point where a dedicated port beats exchange peering, typically the incumbent’s fixed and mobile arms, the other national mobile operators, and sometimes one or two large fixed-broadband challengers. The right list depends on your workload: a gaming platform’s traffic concentrates differently than a video platform’s. Build your list from measured traffic on the exchange, not from a generic template.
Expect PNI negotiations to move at the pace of the largest network involved — plan for a longer runway with the biggest operators and a shorter one with mid-tier networks. The negotiation typically covers port size, redundancy, handover, and SLA — and “redundancy” is a word worth defining explicitly, since foreign operators and Indonesian counterparties don’t always mean the same thing by it.
A pattern that works: activate your domestic IPT on Day 1, join EPIX in Month 1, and at the same time identify the top 10 networks by measured volume. Then, open PNI conversations with that data in hand. Counterparties can already see your traffic on the exchange, which tends to move the conversation faster than a projection would.
Common Mistakes and How to Avoid Them
Over-announcing from the Jakarta node. This pulls in traffic you didn’t intend to serve and inflates pulls back to origin. Announce only what you’re genuinely serving locally, and keep advertisements narrow for Indonesia-resident content.
Still running top-volume traffic over the exchange at month six. If your largest two or three eyeballs still have no PNI by then, you’re pushing bilateral-scale volume through shared infrastructure. Review per-peer traffic quarterly and treat crossing the threshold as a trigger to act, not an agenda item for next year’s planning.
MED problems that only show up after an upstream change. These create intermittent regressions that are hard to diagnose from inside your own PoP. Synthetic probes from Indonesian eyeball networks — not just from the cache itself — are what usually surface this.
Stale PeeringDB or IRR records. These cause requests to stall and routes to get filtered, sometimes months after the fact. Audit both twice a year and after every prefix change.
FAQ
What is EPIX? a carrier-neutral, ultra-high-speed Internet Exchange powered by a high-performance, fully redundant network platform. It enables carriers, ISPs, content providers, and enterprises to exchange IP traffic quickly, easily, and cost-effectively.
Do I still need bilateral PNIs if I’m already on EPIX? Only once a specific eyeball network’s traffic volume justifies a dedicated port. Most regional operators add PNIs based on measured traffic rather than a pre-built list.
How long does it take to set up a PNI with a major Indonesian operator? It depends on the operator and your own negotiating track record — larger networks generally take longer than mid-tier ones.
Where is EPIX physically located? EPIX is hosted inside Digital Edge’s own Jakarta data center facilities – EDGE1 and EDGE2, so operators colocating with Digital Edge reach the exchange over a short in-building cross-connect rather than a cross-metro fiber run.
Conclusion
The peering layer is where a Jakarta cache deployment either earns its keep or quietly underperforms, and the work splits into two phases. Before launch: PeeringDB, IRR, RPKI, and a published policy, this is what determines how fast the market will transact with you. After launch: route server tuning, MED policy, and bilaterals added once traffic is measured rather than guessed at.
Operators who treat the first 30 days as a measurement window, not a launch window, tend to land deployments that match what they modeled beforehand. Operators who invert the sequence spend a quarter negotiating for reach they could have had in a week.
If you’re planning cache capacity in Jakarta, EDGE1 and EDGE2 put you in the same building as EPIX — one less hop, one less dependency, one less thing to negotiate before your traffic is live. We can walk through port options, current route server coverage, and cross-connect logistics before you order anything.
Talk to the EPIX team about what your traffic profile needs on day one.
