Author Topic: Tech Woes  (Read 6721 times)

0 Members and 1 Guest are viewing this topic.

Offline The Latinist

  • Cyber Greasemonkey
  • Technical Administrator
  • Reef Tank Owner
  • *****
  • Posts: 8205
Re: Tech Woes
« Reply #180 on: November 30, 2019, 09:04:09 PM »
So after far too many hours spent troubleshooting, I finally figured out the issue.  Our previous sysadmin had implemented an Airgroup server on our Aruba wireless controller, which worked fine for WiFi but which didn’t allow bonjour or or mDNS between VLANs on our wired network. To fix that, I implemented an mDNS reflection server on our HP switch (the default gateway for the whole network), which worked for the Ethernet but which suddenly made bonjour over the WiFi network stop working. Troubleshooting that was complicated by the fact that I’d made several other configuration changes on the switches at the same time, and so I wasn’t sure which of my changes was causing the issue. Add to that the fact that I wasn’t entirely sure on what switches/ports everything was, and it was a hot mess.

Turns out that implementing mDNS/bonjour for the WiFi on the controller and for the Ethernet on the switches was a bad idea. the two servers conflicted with each other in a way that would probably be perfectly obvious to someone with a better understanding of how mDNS works but which I still don’t quite understand.  Ultimately I fixed the whole thing by turning off the Airgroup server on the Wireless controller entirely and implementing mDNS reflection for both the wireless and wired VLANs on the gateway switch. Everything seems to be working much better now.

Request for advice: This incident has made it clear to me that I need a better understanding of layer 2 and 3 networking. Can anyone recommend a good resource (book, class, or even certification) that would give me a good grounding in such things? The network I’m currently working on is Aruba/HPE, but I’d like to get a basic understanding of Cisco terms for the future, too.
I would like to propose...that...it is undesirable to believe in a proposition when there is no ground whatever for supposing it true. — Bertrand Russell

Offline The Latinist

  • Cyber Greasemonkey
  • Technical Administrator
  • Reef Tank Owner
  • *****
  • Posts: 8205
Re: Tech Woes
« Reply #181 on: December 01, 2019, 08:34:37 PM »
This one is not my own, but my children's school district's. They were hit with ransomware that infected their entire network and encrypted pretty much every workstation, their network drive, and all backups.  They've been without access to their data for 4-6 weeks, and it's looking like they may never get it back.  I don't know how many thousands of dollars they've already spent on it.

It's been a bit of a wakeup call to me, and I'm hoping to have someone in at work to evaluate our backup strategy.  We've got pretty good virus protection through our firewall, but I have no idea whether we've got adequate protection from such an attack and I think we need some real expertise for this one.
I would like to propose...that...it is undesirable to believe in a proposition when there is no ground whatever for supposing it true. — Bertrand Russell

Offline werecow

  • Cryptobovinologist
  • Stopped Going Outside
  • *******
  • Posts: 5294
  • mooh
Re: Tech Woes
« Reply #182 on: December 02, 2019, 01:25:43 PM »
I've used tensorflow before, but always in my own code. Even then I hated how unintuitive it is. Now that I have to plug a (rather complicated) custom neural network into an existing  reinforcement learning framework (stable-baselines), it's pure hell. "Don't be evil"? Yeah, right. It's like they intentionally designed it to make it as hard as possible to debug. The error messages are often vague and sometimes contain no reference to the line of code where things actually go wrong (it just points to some tf function that is called by some other function that I'm calling from a third function), and you can't even print anything until your entire graph works (but to get it working, you often want to see what is actually going on). I tried enabling eager execution, then tested whether it was enabled in the very next line, and it told me "nope". Then I changed one line in a completely different part of the code, and all of a sudden it is enabled and now I get a "eager execution not supported" error. It would be doable if at least keras was supported, but of course keras does not work yet with the stable-baselines library, because apparently there are two different ways of storing a variable, and the one keras uses does not share data between models even when "reuse" is set to true. Apparently this is solved in tf 2.0, but of course this is also not supported yet in stable-baselines.

I'm gonna sit in a corner and shake uncontrollably for a few hours. TTFN!
Mooohn!

Offline stochastibot

  • Off to a Start
  • *
  • Posts: 14
Re: Tech Woes
« Reply #183 on: December 07, 2019, 09:16:49 AM »
Yup, and once you jump to TF2.0, it's time to learn a whole new API.  It was much nicer than Theano when it came out, but it's painful a lot of the time.  Have you tried pyTorch yet? 

Offline werecow

  • Cryptobovinologist
  • Stopped Going Outside
  • *******
  • Posts: 5294
  • mooh
Re: Tech Woes
« Reply #184 on: December 07, 2019, 09:23:36 AM »
Yup, and once you jump to TF2.0, it's time to learn a whole new API.  It was much nicer than Theano when it came out, but it's painful a lot of the time.  Have you tried pyTorch yet?

Yes, and I like it much better, but I'm trying to set up a fairly complex deep RL system using stable-baselines, which is still TF1.x only and apparently not compatible with Keras (or tf.keras) because of the way they share variables between models. They're working on a TF2.0 version, but I'm not very hopeful that it'll be available for the next few months.
Mooohn!

 

personate-rain