How Much Traffic Can Pi-hole Handle?

Announcements
1

Originally published at: How Much Traffic Can Pi-hole Handle? – Pi-hole

We often get questions like: How many clients can Pi-hole handle? Or will Pi-hole slow my network down? To answer the latter, no. Pi-hole actually makes your network faster. To the former, that will vary based on the hardware specs you give to Pi-hole. We have a few real-life examples that can give you some ideas.

Some Background

Pi-hole was designed from the start to be lightweight on resources so it can handle quite a bit. There is often a misconception that all your network traffic flows through Pi-hole, but that is not the case:

Pi-hole is not a proxy.

Pi-hole only handles your DNS traffic, and queries are typically 512 Bytes or less (UDP).

Large Or Enterprise Environments

During the development of FTL, we tested a 4GB RAM VM and it was able to handle well over a million queries in 24 hours. We even cranked it up beyond that.

![|600x103](upload://tG4htCRg22i4mObPekR3NKMEHx7.gif)

Documented Cases

This mid-sized business runs 475 clients on a Raspberry Pi B3 along with Active Directory. The thread has quite a few comments, and it’s an interesting look into Pi-hole’s low resource limit while providing a great benefit.

Another interesting case is one company using Pi-hole on a Raspberry Pi 1 with 50 clients. At the time of the post, they had been running Pi-hole for six months without issue.

It’s also documented that people have gotten anywhere from 30-40 clients and up to 60 (but with a Pi 3).

Some others are less detailed in terms of documentation, but one person has it running with 100 clients. While another person uses a 2GB RAM, 2 CPU VM with 200 clients and it “doesn’t ever make a blip on the server’s resources radar“.

Finally, 400 unique clients can connect to a VM with a single processor, 512MB RAM, but typically around 250 clients are connected at a single time.

See also this Twitter thread for more examples.

This Reddit thread also discusses some numbers.

Undocumented Cases

The highest undocumented case we have heard of was 100-250 clients running on a Raspberry Pi 2.

Small Office / Home Networks

For home users, a Raspberry Pi B works just fine for 10-15+ clients. We even asked you how many devices you had: here were your responses. But if you read the above section, you should feel a confident it can handle your home network, even if it has lot’s of IoT devices in addition to your user devices.

We have heard from many users that they love using Pi-hole on the really old Raspberry Pi’s since it runs great on minimal resources (but don’t forget, you don’t need a Raspberry Pi to run Pi-hole; it can be installed your own Linux hardware).

6 Likes
2

The guy with the Raspi Pi 1 (50 clients) should use another case for the Pi.

The first ones had some problems with flashlights. I think they rebooted (or crashed), when exposed to it.

4

Works fine whit 6k+ clients :+1:

2 Likes
5

What environment? Corporate, Internal? Just curious... What kind of clients do you have?

6

Corporate we run them in VMs

8

Oh, interesting... Do you use configuration management or just set it and forget it? If so which tool?

9

HELLO, I have a community network and 2,500 clients. Does it really work with 6000? What hardware did you use?
Thank you

10

I run pi-hole on a small company network of approximately 15 users / 70 devices reported by our router using a pi zero W ( and yes, no hats, it's on wifi -- performance is fine) powered by the USB connection on the router. I only see load above 0.1 if I'm doing administrative tasks, and the pi runs about 40-43 Celsius. Memory usage runs 10-15%. There were a few teething problems at first, and some downtimes suspected to be caused by the pi-hole, but we worked through those and with 5.1.2/5.1.1/5.2 everything is super stable. Some users need to disable it occasionally to use Google Ads or whatever but for the most part it's not even noticed. I've deployed 6 or 7 (pushed them on friends and family LOL) because the pi zero w was so inexpensive from microcenter for a while and it's such a great addition.

11

Awesome post, thank you for that.

I am wondering about the comparison of the above numbers of clients and ability "to handle traffic"

Does the ability "to handle traffic" depend on
Hardware only or also on the size of gravity DB and FTL databases or this will be already factored in the Hardware choice?

For example people above reported 2500 clients ... 6000 clients but are they using a standard database or added some.

I use Pi Zero W and I have maybe 5-20 devices connected to the PiHole at the same time. My domain count varies about 1-3 million domains. and Gravity DB 100-400 MB

12

Yes. The gravity database size is only proportional to the number of domains on your blocklists. Due to using a balanced tree looking up a domain in millions of domains is only marginally more work than doing this in hundred domains. The FTL database is even entirely optional. Pi-hole will perform exactly equally well without it. You will just loose the ability of long-term data lookups.

The real limitation is mostly on the available memory. CPU doesn't even weigh all that much as Pi-hole is efficient in memory usage. However, keeping the recent 24 hours in memory simply takes the amount of memory this takes. With millions of queries, this will also take some millions of kilobytes (= Gigabytes) of memory. Keeping all this in memory is a limitation we intentionally chose for speed. It should be fair as Pi-hole is so memory-efficient that network with a few hundred thousands of queries per day still fit onto a Zero. If your network is larger than that, you will likely not want to rely with a critical component on a Raspberry Pi and have several GB of memory available.

1 Like
13

Thank you again, makes a lot of sense now.

14

With what kind of setup?

15

Pihole is insanely good at heavy traffic. I have two and due to lots of home automation I’ve shoved into HomeKit I suffer from high dns queries on my network.

image
image1536×2048 458 KB

16

A pi4 is crazy fast for everything but large historical queries. The team has built an insanely good product.

17

Thanks!

We're working on server-side pagination. We hope to get also long-term queries notably faster, however, the main limiting factor here is of course non-sequential disk (SD card) reading speed.

18

The as cards are dog slow, when mine begin to die I’m switching to ssd for my boot drives.

19

This topic was automatically closed after 88 days. New replies are no longer allowed.