> For the complete documentation index, see [llms.txt](https://docs.helodata.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.helodata.com/troubleshooting/slow-speeds.md).

# Slow speeds

"Slow" is rarely the gateway itself — it's almost always one of: the residential / mobile network, your client config, or the target site. Walk through the diagnosis below.

## Baseline expectations

| Product     | First byte | Throughput per connection |
| ----------- | ---------- | ------------------------- |
| Residential | 300–800 ms | 10–50 Mbps                |
| Mobile      | 500 ms–3 s | 5–20 Mbps                 |
| ISP         | 50–150 ms  | 100 Mbps – 1 Gbps         |

If you're seeing this and assuming it should be faster — re-set expectations. Residential is *real* residential bandwidth.

## Step 1 — measure where time goes

```bash
curl -w "\nDNS:%{time_namelookup}s Connect:%{time_connect}s TLS:%{time_appconnect}s TTFB:%{time_starttransfer}s Total:%{time_total}s\n" \
     -o /dev/null -s \
     --proxy http://USER:PASS@gate.helodata.io:7777 \
     https://example.com
```

Output decoded:

| Field                | Meaning                                      |
| -------------------- | -------------------------------------------- |
| `time_connect`       | TCP to **gateway**                           |
| `time_appconnect`    | TLS handshake to **target** (through tunnel) |
| `time_starttransfer` | First byte from target                       |
| `time_total`         | End to end                                   |

Where is the time going?

* High `time_connect` (>500 ms) → network from you to `gate.helodata.io`. Switch to a regional gateway (`na.gate.helodata.io`, `eu.…`, `ap.…`).
* High `time_appconnect` → target is slow or far from the exit IP.
* High `time_starttransfer` after handshake → target server processing, not network.

## Step 2 — eliminate connection re-establishment

Re-using TCP / TLS connections (HTTP keep-alive, HTTP/2) saves the handshake cost. In `requests`:

```python
s = requests.Session()                   # persistent connection
```

In Node with undici, the default dispatcher pools connections automatically — just don't construct a new `ProxyAgent` per request.

## Step 3 — check your concurrency

You might have hit a per-sub-user concurrent limit:

```
HTTP/1.1 429
X-Helodata-Error-Code: concurrent-limit
```

See [Rate limits](/api-reference/rate-limits.md). Either lower your concurrency or upgrade your plan.

## Step 4 — try a different exit IP

The selected residential IP might just be slow today (it's someone's home internet). Drop your `session` to get a different IP.

## Step 5 — try a different product

If residential is too slow for your throughput needs, ISP gives you 10–50× the per-IP throughput at a different cost model.

## When the *gateway* is slow

Rare but real. Check the status page:

* <https://status.helodata.com>

If status is green but you're seeing high `time_connect` consistently, open a ticket with `traceroute gate.helodata.io` output and `X-Helodata-Request-Id` from a slow request.

## Common pitfalls

* **SOCKS5 with local DNS** — local resolver adds latency. Use `socks5h://`.
* **TLS-wrapped connection to gateway** — adds an extra handshake on top of the target's own TLS; only use when your network demands TLS-only egress.
* **DNS cache cold** — `time_namelookup` >100 ms means your DNS resolver isn't caching. Configure a local resolver.
* **Logging slows your client** — verbose logs can dominate the wall-clock time on small requests.

## Benchmarking properly

To compare proxy speed, measure many requests, not one:

```bash
for i in $(seq 100); do
  curl -w "%{time_total}\n" -o /dev/null -s \
       --proxy http://USER:PASS@gate.helodata.io:7777 \
       https://example.com/sitemap.xml
done | sort -n | awk 'NR==50{print "median:",$1} END{print "max:",$1}'
```

Median > 95th percentile — the latter is where bot detection lives.