benchess

6 ms·

1.
by benchess 11mo ago
> - It's not possible to set screen time restrictions for Safari. I thought this too. I discovered it actually is possible though, just doesn't appear in the list. Go "Screen Time" -> "App Limits" -> &
2.
Kubernetes with 1M Nodes(github.com)
14 points by benchess 11mo ago|1 comments
3.
by benchess 11mo ago
Heya, wanted to show off a project of mine to identify the scaling limits of a large Kubernetes deployment
4.
by benchess 1y ago
The irony of posting this on GitHub which remains shamefully without IPv6
5.
by benchess 2y ago
It’s mentioned
6.
by benchess 2y ago
My kids like reading this over and over: Now & Ben: The Modern Inventions of Benjamin Franklin: https://a.co/d/jhm2SvM
7.
by benchess 2y ago
Use a dns proxy and https://github.com/froggeric/DNS-blocklists/blob/main/NoAppl... You may also need to disable Private Relay
8.
by benchess 3y ago
They’re for sale on Mouser for $20625 each https://www.mouser.com/ProductDetail/BittWare/RS-GQ-GC1-0109... At that price 568 chips would be $11.7M
9.
by benchess 3y ago
This isn't running on one chip. It's running on 128, or two racks worth of their kit. https://news.ycombinator.com/item?id=38739106 This doesn't mean much without comparing $ or watts of GPU equivalents
10.
by benchess 3y ago
If I understand correctly, Groq chips have 220MB SRAM and the next best level is DDR4? How many chips are needed to run Llama2-70B at those speeds?
11.
by benchess 5y ago
The key constraint here is that the author has no access to persistent disks and can only use object storage for persistence. Otherwise Thanos would be extreme overkill for this number of metrics. Single-node VictoriaMetrics can easily hand
12.
by benchess 6y ago
Hi, co-author here. Yes we are excited about the potential of this!
13.
by benchess 6y ago
Hi, co-author here! We use a pretty standard tech stack of PyTorch + NCCL + MPI. We've used both OpenMPI and MPICH to varying degrees. Kubeflow is interesting, but it solves a slightly different problem of scheduling/coordinating
14.
by benchess 6y ago
Hi! Co-author here. We do keep the nodes running 24/7, so Kubernetes still provides the scheduling to decide which nodes are free or not at any given time. Generally starting a container on a pre-warmed node is still much much faster t
15.
by benchess 6y ago
It's all pretty well covered in the Methodology section of the about page: https://takecongress.org/about.html
16.
by benchess 7y ago
Utilize SNI and serve up a fake cert when someone scans you without a matching hostname. Censys is scraping you by IP, so it'll just see the fake cert. You can do this in nginx by making the fake cert the first server block.
17.
by benchess 9y ago
Not all bandwidth is created equal. Different providers can have widely different levels of performance. Higher-priced bandwidth _can_ also be better bandwidth. Cloud providers haven't done a great job of explaining this so far, but Go
18.
by benchess 11y ago
What an amazing story. Great to hear of people like Katherine!
19.
by benchess 11y ago
So when you hit the limit you want AWS to stop your spending... how? You're spending $/hr on compute, so terminate your instances. Plus your EBS volumes, so delete your drives. Plus... S3? So delete all your data?
20.
by benchess 12y ago
Also Stratos https://stratoscard.com/
21.
by benchess 12y ago
"No planned downtime: Local SSD data will not be lost when Google does datacenter maintenance." Only unplanned downtime then :) Live migration maybe adds another '9' to reliability, but customers still need to plan for f
22.
by benchess 12y ago
It's both a product and a platform. plus.google.com and its "What's New?"-style newsfeed is a tangible product. The news that Google+ integration is no longer required for other Google products means the platform is depr