# 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.


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.helodata.com/troubleshooting/slow-speeds.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.