Category: Uncategorized

When we talk about IP protocols and IPv6 in this particular case we think about routing over the internet. But what many do not know is that IPv6 is already playing a major role in all kinds of systems.

In this post I’m not going to explain IPv6 in detail, for many things I’ll link to external pages.

Ad-Hoc networking using Link-Local

IPv6 has a great feature where each host on a network will generate a Link Local Address (LL) which is mandatory according to the protocol.

The great thing about LL is that it allows hosts to establish communication without the need of DHCP or anything else. Just plug in a host on a Layer 2 Ethernet network and they can start communicating right away.

fe80::5054:ff:fe98:aaee is an example of a IPv6 LL address and using this address it can search for (via multicast) and communicate with other hosts in the network.

Examples of systems using IPv6 LL are Matter and Combined Charging System (CCS).

Combined Charging System (CCS)

CCS is a standard for (fast charging) electric vehicles. Some years ago I stumbled upon multiple documents showing that CCS would be using a combination of PLC, IPv6, TCP/IP, UDP, HTTP, TLS and many other very common protocols for communication.

Using PowerLine Communication (PLC) the vehicle and charger will establish a Layer 2 Ethernet network over which they will communicate using IPv6 LL.

I was not able to find the exact details of the IPv6 part of the CCS standard, but it is very interesting to see that IPv6 is being used for something like charging electric vehicles.

Kudos to the people behind CCS and for using IPv6 for this!

In my opinion it’s a no-brainer to use IPv6 for such functionality as the LL addressing allow you to create networks very quickly and in a reliable manner.

You can also use a very well documented protocol and the many tools for IPv6.

Now I would like to run Wireshark on such a link!

If you have more information about this, please contact me!

It’s been almost a year that I’ve been living in my new house and proud father of a son.

Almost every parent wants to watch their little on through a camera to see them sleeping in their bed.

There are many, many, many baby monitors on the market but in my experience they are all crap. Our house is build with a lot of concrete and none of them was able to penetrate the walls in our house and thus have a reliable video connection.

I also tried some IP/WiFi based baby monitors from for example Foscam, but these are all cheap Chinese cameras and didn’t work either. Crashing iPad apps, sending random (UDP) packets to China, half-baked UIs, etc, etc. They were all very low quality.

After a lot of frustrating I bought a Unifi Video G3 (UVC-G3-BULLET) camera and mounted it on the bed of my son.

I am using multiple Unifi video cameras around my house using Unifi Video, so for me it was just a matter of adding an additional camera to my Unifi system.

RTMP + HLS with Nginx

I also experimented with building a video stream using RTMP, ffmpeg and HLS with Nginx.

The Unifi cameras can export a RTSP stream which you can set up to live stream with Nginx. I tried this and it works for me, but with a delay of ~20s I thought it was not usable on my use-case.

It does however allow for a very easy way of building your own webcam!

In the past years I’ve said goodbye to hardware RAID controllers and mainly relied on software solutions like mdadm, LVM, Ceph and ZFS(-on-Linux) for keeping data safe.

At PCextreme we use hypervisors with local NVMe storage running in Linux’s mdadm software RAID-10. This works great! But I wasn’t satisfied with the performance for a few reasons:

• It is expensive on the CPUs (Dual AMD Epyc 48-core)
• It’s not super fast

We mainly use the Samsumg PM983 (1.92TB) devices and I started to look around if there is a hardware solution which could offload the RAID computation to a dedicated SoC so it wouldn’t eat up our CPU cycles.

After searching I found the Broadcom SAS3916 chip which is on the MegaRAID 9516-16i controller from Broadcom. This chipset supports NVMe devices in various RAID modes.

I wanted to benchmark Linux’s software RAID against the Broadcom controller to see if it would be faster and save us the expensive CPU cycles.

With mdadm we also looked into RAID-5/6 to have more usable space. We however found out that this eats up so many CPU cycles that it wasn’t feasible to use in production for our purposes.

Benchmarking setup

• Ubuntu Linux 18.04 with kernel 5.3
• SuperMicro 1114S-WN10RT
• AMD Epyc 7302P 16-core CPU
• 128GB Memory
• 4x Samsung PM983 1.92TB
• Broadcom MegaRAID 9516-i

Benchmarking will be done using fio and the main elements we are looking for:

• QD=1, 4, 8 and 16 4k random write performance
• CPU utilization during benchmarks

MegaRAID configuration

The RAID-10 array was created using storcli

storcli64 /c0 add vd type=raid10 drives=252:4,6,8,10 pdperarray=2

mdadadm setup

RAID-10 was set-up using this command:

mdadm --create --level=10 --raid-devices=4 /dev/md0 /dev/nvme0n1 /dev/nvme1n1 /dev/nvme2n1 /dev/nvme3n1

fio

The fio jobs we ran to benchmark:

[global]
ioengine=libaio
direct=1
invalidate=1
bs=4k
runtime=300
filename=/var/lib/libvirt/images/fio
size=64g
rw=randwrite
numjobs=1

[rw_mdadm_1]
iodepth=1

[rw_mdadm_4]
iodepth=4

[rw_mdadm_8]
iodepth=8

[rw_mdadm_16]
iodepth=16

[rw_mdadm_32]
iodepth=32

Results

Short answer? The MegaRAID controller is much faster in RAID-10 mode and also saves up to 30% of CPU cycles on the Linux machine.

You can also download the JSON output from fio here:

• mdadm results
• Broadcom MegaRAID results

In addition you can download my Open Office spreadsheet I used to generate the graphs and process the data.

Graphs

Some graphs with the results of the IOps and CPU utilization.