Skip to content
streamarsRequest early access

Same hardware. Same streams. Measured.

We ran streamars and XC side by side on identical servers with the same channels, the same catalog and the same load. These are the medians of three runs.

Measured October 2026.

  • API latency p99

    100 requests/s, 20 concurrent

    8.3× lower (lower is better)

    streamars
    4.9 ms
    XC
    40.6 ms

  • API CPU, average

    100 requests/s, 20 concurrent

    3.4× lower (lower is better)

    streamars
    33%
    XC
    113%

  • TS first byte p99

    100 → 300 → 500 viewers

    35% lower (lower is better)

    streamars
    12–17 ms
    XC
    26–58 ms

  • TS CPU, peak

    100 → 300 → 500 viewers

    2.2× lower (lower is better)

    streamars
    142–156%
    XC
    336%

  • TS memory

    100 → 300 → 500 viewers

    3× lower (lower is better)

    streamars
    ~1.0 GB
    XC
    ~3.0 GB

  • HLS first byte p99

    100 → 300 viewers

    44% lower (lower is better)

    streamars
    9–11 ms
    XC
    19.5–21 ms

  • Channel switch, first byte p99

    50 viewers zapping every 10 s

    2.9× lower (lower is better)

    streamars
    5.9–6.9 ms
    XC
    20–23.5 ms

  • Idle CPU

    10 channels, no viewers

    2.1× lower (lower is better)

    streamars
    17.7%
    XC
    37%

  • Idle memory

    10 channels, no viewers

    45% lower (lower is better)

    streamars
    948 MB
    XC
    1,711 MB
Show as a table
MetricScenariostreamarsXC
API latency p99100 requests/s, 20 concurrent4.9 ms40.6 ms
API CPU, average100 requests/s, 20 concurrent33%113%
TS first byte p99100 → 300 → 500 viewers12–17 ms26–58 ms
TS CPU, peak100 → 300 → 500 viewers142–156%336%
TS memory100 → 300 → 500 viewers~1.0 GB~3.0 GB
HLS first byte p99100 → 300 viewers9–11 ms19.5–21 ms
Channel switch, first byte p9950 viewers zapping every 10 s5.9–6.9 ms20–23.5 ms
Idle CPU10 channels, no viewers17.7%37%
Idle memory10 channels, no viewers948 MB1,711 MB

How we measured

  • Two identical cloud servers (4 vCPU, 8 GB) in the same datacenter, one panel under load at a time.
  • A separate machine ran the load generator and a test origin: 10 live channels of 720p H.264 at 2.5 Mbit/s.
  • The same VOD catalog on both: 2,002 movies and 200 series served from local disk.
  • Three runs per scenario, alternating panels. Request-rate limits were lifted on both panels for the load generator.
  • Where a result varied across runs, the bars use the run least favourable to streamars.

What to make of it

Both panels served every faststart movie and every live viewer without errors in the main scenarios. The gap is in how much of the server each one needs to do it, how fast playback starts, and what happens at the edges: seeking, non-faststart files and catalog listing under load.

Run your own numbers. Early-access operators get a copy of the load generator and the scenarios.

Run your panel on streamars.

We are onboarding operators one at a time and helping each move their XC panel across. Tell us about your setup and we will get you running.

Request early access