Andrew Zonenberg

3.3K Followers
479 Following
29.6K Posts

Security and open source at the hardware/software interface. Embedded sec @ IOActive. Lead dev of ngscopeclient/libscopehal. GHz probe designer. Open source networking hardware. "So others may live"

Toots searchable on tootfinder.

ngscopeclienthttps://www.ngscopeclient.org/
Bloghttps://serd.es
LocationSeattle area
GitHubhttps://github.com/azonenberg

Starting to dust off the Ethernet switch project and make some strides towards finishing it.

First step: preparing to populate the back side of the remaining switch line cards. I hand assembled one a year or so ago and then the project got put on ice while I got busy with higher priority stuff.

But now that my friends are starting to get their PnP operational, I'm gonna try and run the rest of the boards on it.

Just inventoried components for the back-side assembly run and ordered a reel of 0402 zero-ohms because I seem to have misplaced my current one (all I have are the loose ones in a pill organizer from when I was starting out, I swear I had a few hundred on cut tape but they're not where they should be).

I ordered five PCBs originally and populated one by hand, so I have four left over. The plan is to populate three fully since I need a total of four (two 24-port line cards per 48 port switch, two switches).

And since the marginal cost of populating an extra board on the PnP is pretty low, I'm gonna stuff one more but leave all the expensive connectors and semiconductors off. I want to do some 2-port shunt-through VNA measurements on the PDN and get some real world impedance measurements. Not that it's necessary, the design is already brought up and working, but it'll be a good test and validation for the future.

Just pulled a now empty PDU and tidied up a few cables. Main IT rack is looking pretty good now.

Everything below the 1GbaseT switches (except the PoE injector) is pulling a combined total of just under 1 kW; the rest isn't metered yet.

That's somewhat approximate because I don't know how accurate the 330W baseline number is (I've seen as low as 309W during idle-ish conditions) but I'm confident in saying each build costs me <2 cents in electricity.
All done. Still more room for a single slot AMD card in the future.

VM server is unracked again for hopefully the last round of hardware upgrades for a while.

Swapping the unstable 100G NIC for one that's hopefully happy at pcie gen4 speeds, swapping the dual slot GTX 1650 for a third dual slot RTX 3050, and moving to a new PDU.

Somebody asked me to decap something off this board. I assume the main ESS AudioDrive chip?

Is there anything else worth saving or looking at on the board once I've pulled the chip?

I've been banging my head on this for ages and I can't get the Korok to come out. What am I missing?
Halp i got my shonk wet and it's not fuzzy anymore

So far, I've accounted for about 670W of the ~4 kW base load power the lab draws at the UPS, or about 17%.

Here it is broken down by category:
* Storage - the three Ceph cluster nodes
* Networking - core router and one of the 40/100G switches
* Backup - my friend's offsite backup server I host

And the UPS power is only 63% of the building-wide power consumption (average 6.3 kW). I have no data so far into where that 2.3 kW is going - some is normal house stuff like cooking, lighting, HVAC, the living room TV, etc. But a nontrivial fraction is also going to lab overhead like lighting and cooling and unless I get a current transformer on the minisplit feed I won't have visibility into that.

I'm missing data from the virtualization server, the single biggest power hog on the rack based on how much hot air comes out the back, since I don't want to shut down all of my production VMs needlessly. I'd rather wait until I have another reason to shut it down (like installing the third RTX 3050 I plan to buy in the not too distant future).

And this is only data from the main IT rack - I'm missing numbers from all of the workstations, T&M equipment, and non-UPS loads like cooling and lighting.