02:54 <mtrbot-ml> [mattermost] <sb10q> fwiw the linux driver on red pitaya does 56MB/s download and 25MB/s upload, tested with commands like `cat /dev/zero | pv | nc 192.168.1.190 4444`

02:54 <mtrbot-ml> [mattermost] <sb10q> and `nc -l -p 4444 > /dev/null`

03:41 <sb0> rjo: when measuring phases using the technique described in "Oscillator metrology with software defined radio", how sensitive is it to the input signals not being sine waves?

08:26 <rjo> not at all as long as they are rejected by the analog and digital filters

08:28 <rjo> and even if they are in passband or aliased, it would only matter when the higher harmonics are noisier than the fundamental or if there is a non-linearity in receiver, which is both hard to imagine.

12:46 <airwoodix> rjo: did other groups express interest in artiq#1345? or in compression?

13:04 <rjo> airwoodix: I don't think so. There is also #505 with more feature requests on the hdf5 processing. And in our experiments I'd really a feature to suppress the saving of the hdf5 file alltogether.

13:50 <mtrbot-ml> [mattermost] <hartytp> anyone know of PoE switches/injectors/splitters with a decent remote interface to switch the power on/off?

13:58 <airwoodix> rjo: thanks. #505 is pretty old... What format would you use if not hdf5? single .npy/.npz files? plain-text?

13:59 <airwoodix> I wanted to find some time to look into very minimal compression support in artiq. Mainly exposing the arguments to create_dataset in worker_impl. The problem is that the efficient filters are not part of the standard hdf5 distribution, so it adds a pretty annoying dependency (not sure how hard it gets on windows)

14:23 <rjo> airwoodix: depends. There are lots of cases where influxdb is nicer than hdf5 (especially for temporal data, measurements under supposedly constant conditions).

14:25 <rjo> airwoodix: For images I still like png/jpg with metadata loaded into them (xmp or even within the png standard) and lossless compression if desired. We had that at ETH.

14:26 <rjo> for hdr (float64) images probably npz. how much compression do you need and how much do npz, lossless jpeg/png or the various hdr formats give you?

14:30 <rjo> hartytp: the netgear prosafe (gs110tp or gs724tpp) work well. all of the fs.com poe switches i had have failed. the cisco catalyst are also good and not to expensive. all are remote controllable.

14:34 <mtrbot-ml> [mattermost] <hartytp> @rjo thanks! Is there an SDK for the prosafe switches? Or is it just a web interface?

14:37 <rjo> Telnet. See the kasli-i2c repo for a python script.

14:38 <mtrbot-ml> [mattermost] <hartytp> nice. Thanks!

14:42 <rjo> Snmp might also work

14:44 <mtrbot-ml> [mattermost] <hartytp> Is there a reference manual for the telnet interface for the gs110tp?

14:44 <mtrbot-ml> [mattermost] <hartytp> I only found https://www.downloads.netgear.com/docs/m4100/enu/202-11161-01/cli.pdf on a quick skim although it looks like the commands probably work for other switches as well

14:51 <rjo> I don't remember what I used. All the broadcom fastpath based switches might be more or less compatible.

14:53 <rjo> Finding comprehensive Cisco cli docs that are complete, concise, and applicable is similarly difficult.

14:54 <rjo> There is also the gs110tpp with Poe+ now.

14:54 <rjo> But no SFP

14:56 <rjo> The Cisco catalyst seemed to be the best if you need ~24 ports, fanless and Poe+ at reasonable cost.

15:17 <airwoodix> rjo: npz (compressed) is roughly equivalent to gzip which would be sufficient. JPEG-LS is in principle nice as well for up to 16-bit (integer) deep images. Didn't test JPEG-XT. Influxdb has the disadvantage to need to setup a server. IMHO, single files on git (maybe with git lfs) is easier to manage over the years.

15:30 <rjo> And images in influxdb is not what I would do either. It doesn't mesh with the other tooling. Png also works for u16. For both png and jpeg you can store multiple images in a single file. Sure. For images a flat filesystem is just fine. Easy backup, easy access, fast lookup.

16:44 <mtrbot-ml> [mattermost] <hartytp> thanks for all the info @rjo

16:44 <mtrbot-ml> [mattermost] <hartytp> much appreciated