Sardius Entrypoint
As MSL4 retires, what reliable, flexible, and cost-effective solutions are available to become a single endpoint with a single URL?
Sardius Entrypoint fits inside the workflow you already run. Same push, same URL, but behind it a cost-effective technology solution that writes to multiple S3-compatible storage origins, a multi-CDN approach to distribution, and part of an ecosystem that offers multiple flexible solutions to all your clients.
Akamai MSL has served many clients well over many years, but in 2019, Sardius needed alternatives to give our clients greater flexibility and cost options. Sardius Entrypoint has been serving those clients ever since.
An HTTP POST of ordinary 2–10 second segments and M3U8 manifests. If your encoder can push HLS, it can push to us.
A single push lands in several S3-compatible buckets across clouds and regions. No individual origin is a dependency.
Delivery queries those origins in parallel and takes the first correct response, so a slow path simply loses instead of having to fail.
There are only four moving parts, and none of them are complicated. The interesting decision is at the end: we don't pick an origin and fail over when it breaks. We ask several at once and take whichever answers first.
For segments, the first origin to return a byte wins. For live manifests, speed alone isn't enough. A fast but stale origin would rewind the live edge, so the race waits briefly for a second opinion and serves the freshest manifest. Nothing is marked unhealthy and nothing is switched over; a losing origin just loses.
Standard HLS push. An ordinary HTTP POST of 2–10 second segments and their M3U8 manifests, to a publish URL with a token. Point a second encoder at the backup endpoint and you have failover. No SDK, no proprietary protocol.
POST /live/{stream}/index.m3u8
Each segment is written to several S3-compatible origins simultaneously, across different providers and different regions. One push becomes many durable copies.
→ storage 1 · 2 · 3 · yours
On the way out, the request is put to those origins at once. Segments take the first byte back. Live manifests wait a beat for a second opinion and take the freshest, so the live edge never rewinds.
edge ⇄ origin · origin · origin
Standard HLS out, TS or CMAF/fMP4, from whichever path won. Ordinary players, ordinary manifests, no client-side integration.
GET /live/{stream}/index.m3u8
Failover is reactive. Something has to notice a problem, agree it's real, and then move traffic, and every one of those steps costs seconds while viewers are already buffering. Racing removes the detection step entirely. A slow bucket doesn't need to be declared unhealthy; it just loses the race and nobody notices.
The winner isn't a decision made at configuration time either, and it isn't decided once per stream. The race is per file: every manifest fetch and every new segment is raced on its own, so the origin serving a viewer reflects conditions in the last second or two, in their region. Nothing is pinned for the duration of a broadcast.
Health checks mark an origin down. DNS or configuration moves traffic to a standby. Viewers absorb the gap while all of that happens.
Recovery measured in seconds to minutes
The request goes to every origin at once and the winner answers. A degraded origin loses races instead of triggering an incident, and each edge re-races independently.
Recovery measured in one race, at the edge
Redundancy that doesn't depend on anything noticing a failure, and global delivery that isn't tied to one CDN's bad day or one storage provider's slow region.
Akamai MSL4 reaches end-of-life on December 31, 2026, and Akamai's recommended replacement, MSL5, is built and owned by a separate video infrastructure company, not by Akamai itself. Adding Entrypoint as a destination is one more standard HLS push. No new architecture, just seven years in production, battle-tested.
Akamai deactivates MSL4 streams on or soon after December 31, 2026. Every MSL4 customer has to choose a destination between now and then.
Once a customer touches MSL5 settings or Harmonic ingest hostnames, Akamai states there's no rollback. The customers deciding right now are the ones worth reaching first.
Feature comparison
Side by side, feature by feature.
| Criterion | Sardius Entrypoint+ raceworker delivery | Akamai MSL4Retiring Dec 31, 2026 | Akamai MSL5Powered by Harmonic |
|---|---|---|---|
| Status | In production since 2019, actively developed | End of life Streams deactivated on or soon after Dec 31, 2026 | Generally available; migration window runs through 2026 |
| Engine ownership | Sardius Media. Neutral, independent infrastructure vendor | Akamai | MediaKind, as of June 2026. Formerly Harmonic's video business; Harmonic is now pure-play broadband and no longer owns the technology |
| Migration path | Add one HLS push destination. Reversible at any time | n/a (this is the thing being left) | One-way. Akamai states no rollback once MSL5 settings or Harmonic hostnames are touched |
| Role | Live origin and delivery, with manifest assembly on the way out | Live origin + packager | Live origin + packager |
| Ingest | HTTP POST HLS push with token auth (PUT also accepted). Standard, known file structure | HTTP push, HLS and DASH | HTTP POST, digest or header-key auth, IP allowlist, primary + backup |
| Packaging out | HLS: TS and CMAF/fMP4 | HLS, DASH | HLS, DASH, CMAF |
| HEVC | Shipped and tested against real AWS Elemental HEVC traffic | Supported | Supported |
| DVR window | No architectural cap. dvrLength is set per stream on the publish token, bounded by storage. 24/7 streams roll on uninterrupted, but make a new archive every 24 hours |
30 minutes sliding. Instant TV extends playback to the last 12 hours, but requires Advanced Archive | 12-hour sliding playlist maximum |
| Archive / live-to-VOD | Automatic the moment a stream ends. Keep indefinitely, or expire on a schedule you set | Basic Archive (indefinite, or purged on a policy) or Advanced Archive (up to 31 days, required for Instant TV), with manual clip-out of the time range | Live2VOD clips up to 24 hours long; archive retention up to 62 days |
| Scheduled go-live | Supported on a channel with a calendar event. Start the encoder early and rehearse against a private preview playlist, then the public stream begins at the scheduled time | Not documented. MSL4's startTime is an Instant TV playback offset, not a deferred public start |
Not documented |
| Storage | Bring your own: any S3-compatible bucket. Or our multi-cloud pool, or both at once | Akamai NetStorage | Akamai-internal |
| CDN | Multi-CDN: Cloudflare and Fastly, both offered in the master manifest and both running the same delivery code | Akamai | Akamai |
| Ingest regions | Cloudflare global edge; storage spread multi-region | Akamai network | Three regions: SEA, ORD, LON |
| Live manifest control | Live clipping to VOD, DVR resized on the fly, multi-language VTT injection, audio/subtitle track filtering, embedded CEA-608 declaration, stream delay, audio-only renditions, trick-play thumbnails, ad-break signalling | Not offered | Not offered |
| Provisioning | One API call returns the stream and its publish token | Stream configuration in the Akamai portal | Stream config plus CNAME / AMD property setup |
| Health & monitoring | Per-variant status JSON, alerting, 5-minute idle auto-end | Portal + reporting | "Self-healing network" claims; dashboard and API |
| Primary / backup ingest | Supported. Primary and backup publish URLs, with failover between them | Supported | Required, with primary and backup in different locations |
| Geo restrictions | Enforced at delivery today. Per-stream country and region allow/deny lists, applied to live and archives. Fastly parity is Early Q4 | Via separate Akamai products | Via separate Akamai products |
| DASH output | On request Same logic, different container | Yes | Yes |
| Ad signalling | Delivery-side ad-break signalling ships today. EXT-X-DATERANGE and SCTE-35 interop markers projected into live and VOD manifests. Encoder-inserted SCTE-35 passthrough is Early Q4 |
Not documented | Not documented |
| Content protection | Token auth on the publish side, geo restrictions enforced at delivery, and viewer-facing DRM today via EZDRM through our Bitmovin player | Not documented | Not documented |
| Captions | Multi-language WebVTT injected at delivery, with subtitle playlists built on the fly, default-track selection, and embedded CEA-608 declared in the master playlist | WebVTT | WebVTT, IMSC1 / SMPTE-TT |
| Low-latency HLS | Queued | No | Beta, synthesized from standard HLS, 4–6s claimed |
| Transcoding | None by design; we take what your encoder sends | None | None today; roadmap / separate product |
MSL4 and MSL5 rows reflect Akamai's published documentation as of September 2026. "Not documented" means we could not verify the capability in Akamai's public materials, not that it's absent. Items marked Early Q4, Queued or On request are covered in the roadmap below.
Raceworker is what sits in front of the origins. It's the thing running the race, and because every manifest passes through it, it's also the place a live stream can be reshaped without going anywhere near the encoder.
Every origin is queried in parallel, per file. Segments take the first byte returned; live manifests are compared on write time so the freshest wins. A new segment is a new race, so nothing gets pinned to one origin mid-broadcast.
Pull a section straight out of a running live manifest and serve it as a VOD while the stream is still going. A start point and a duration, no re-encode and no export step.
Shrink a 12-hour rolling window to five minutes mid-stream, or set a different length per stream. The window is a runtime value, not a provisioning decision.
Multi-language WebVTT injected at delivery, with subtitle playlists built on the fly, a chosen default track, and embedded CEA-608 tracks declared in the master playlist. Shipped in production for live captioning.
Filter a manifest down to specific audio or subtitle languages, isolate a single locale, or serve an audio-only rendition, all from the same stream and without a second encode.
Per-variant status JSON, alerting into ops channels, automatic stream end after five idle minutes. Problems get routed around before they get escalated.
Every one of these features exists because a real event needed it. Live captions on a broadcast. A DVR that had to shrink mid-stream. A storage provider that got slow during a keynote. The delivery layer is shaped by problems we had to solve while the stream was still running.
Bring Your Own Storage
Add your keys to an account and Entrypoint treats your bucket as an origin. Your live segments land in your storage, and we deliver from it. Ours, yours, or both at the same time; it doesn't matter to the pipeline.
The default. Segments fan out across three independent Sardius storage origins with redundancy and racing already wired up.
Any S3-compatible endpoint. Live streams write straight into your bucket, and delivery reads back out of it.
Write to our pool and yours simultaneously. Delivery races across the whole set, so a bad origin never becomes a bad stream.
Entrypoint is battle-tested, and still being actively improved. Here's what's shipping in Q4 2026: what's on the bench, what follows it, and what we'll build on request.
Active engineering work, closing MSL feature parity.
Scoped and sequenced behind the work above.
Not built yet, but available as a custom build. Tell us which of these you need and we'll prioritize it.
Entrypoint is one piece of a larger platform. Sardius Media serves enterprise broadcasters end to end: live streaming, on-demand video, and the infrastructure around both. Whatever the requirement, we can usually supply the answer.