- 168
- 294 147
Oxide Computer Company
United States
Приєднався 10 тра 2021
The only Cloud Computer you can own.
Oxide and Friends 12/16/2024 -- Scaling Bluesky with Paul Frazee
Paul Frazee joins Bryan, Adam, and the Oxide Friends to talk about the inner workings of Bluesky and the AT Protocol. Paul and the Bluesky team have been working on decentralized systems for years and years--very cool to see both the next evolutionary step in those ideas and their successful application in Bluesky!
Notes: github.com/oxidecomputer/oxide-and-friends/blob/master/2024_12_16.md
Context: bsky.app/profile/bcantrill.bsky.social/post/3ldgtnaftkc2m
Notes: github.com/oxidecomputer/oxide-and-friends/blob/master/2024_12_16.md
Context: bsky.app/profile/bcantrill.bsky.social/post/3ldgtnaftkc2m
Переглядів: 1 086
Відео
dtrace.conf(24) Robert Mustacchi - Getting Elves to Talk to Dwarves
Переглядів 102День тому
dtrace.conf(24) Robert Mustacchi - Getting Elves to Talk to Dwarves
dtrace.conf(24) Kris Van Hees - eBPF + DTrace
Переглядів 111День тому
dtrace.conf(24) Kris Van Hees - eBPF DTrace
dtrace.conf(24) Bryan Cantrill & Adam Leventhal - Wrapping up
Переглядів 346День тому
dtrace.conf(24) Bryan Cantrill & Adam Leventhal - Wrapping up
dtrace.conf(24) Aapo Alasuutari - USDTs the Wrong Way
Переглядів 52День тому
dtrace.conf(24) Aapo Alasuutari - USDTs the Wrong Way
dtrace.conf(24) Eliza Weisman - DTracing Two great tastes…better together
Переглядів 82День тому
dtrace.conf(24) Eliza Weisman - DTracing Two great tastes…better together
dtrace.conf(24) Alan Hanson - Building a Monitoring Tool
Переглядів 81День тому
dtrace.conf(24) Alan Hanson - Building a Monitoring Tool
dtrace.conf(24) Dave Pacheco - New User Resources
Переглядів 78День тому
dtrace.conf(24) Dave Pacheco - New User Resources
drace.conf(24) Nahum Shalman - OpenTelemetry Tracing for Dropshot
Переглядів 78День тому
drace.conf(24) Nahum Shalman - OpenTelemetry Tracing for Dropshot
dtrace.conf(24) Luqman Aden - Debugging SDT
Переглядів 34День тому
dtrace.conf(24) Luqman Aden - Debugging SDT
dtrace.conf(24) Bryan Cantrill - State Of The Union
Переглядів 266День тому
dtrace.conf(24) Bryan Cantrill - State Of The Union
dtrace.conf(24) Ben Naecker & Adam Leventhal - USDT Past, Present, and Future
Переглядів 96День тому
dtrace.conf(24) Ben Naecker & Adam Leventhal - USDT Past, Present, and Future
dtrace.conf(24) Kyle Simpson - DError Banishing Debug fmt for nested enums
Переглядів 56День тому
dtrace.conf(24) Kyle Simpson - DError Banishing Debug fmt for nested enums
dtrace.conf(24) Ryan Goodfellow - DTrace + P4
Переглядів 140День тому
dtrace.conf(24) Ryan Goodfellow - DTrace P4
Oxide and Friends 12/9/2024 -- Conferences in Tech
Переглядів 878День тому
Bryan and Adam were joined by Theo Schlossnagle, KellyAnn Fitzpatrick, and Steve O'Grady to talk about conferences in tech. A lot has changed in the past couple of decades about the impetus for conferences and what makes it worthwhile to attend. Notes: github.com/oxidecomputer/oxide-and-friends/blob/master/2024_12_09.md Context: bsky.app/profile/bcantrill.bsky.social/post/3lcv7cd2iz22t
Oxide and Friends 12/2/2024 -- Intel after Gelsinger
Переглядів 1,8 тис.14 днів тому
Oxide and Friends 12/2/2024 Intel after Gelsinger
Oxide and Friends 11/18/2024 -- Technical Blogging
Переглядів 1,1 тис.Місяць тому
Oxide and Friends 11/18/2024 Technical Blogging
Data Center Power Efficiency: A Conversation
Переглядів 762Місяць тому
Data Center Power Efficiency: A Conversation
Oxide and Friends 10/28/2024 -- Books in the Box IV
Переглядів 1 тис.Місяць тому
Oxide and Friends 10/28/2024 Books in the Box IV
Oxide and Friends 10/14/2024 -- Unshrouding Turin (or Benvenuto a Torino)
Переглядів 1,3 тис.2 місяці тому
Oxide and Friends 10/14/2024 Unshrouding Turin (or Benvenuto a Torino)
Oxide and Friends 9/30/2024 -- Querying Metrics with OxQL
Переглядів 2,3 тис.2 місяці тому
Oxide and Friends 9/30/2024 Querying Metrics with OxQL
Oxide and Friends 9/23/2024 -- RTO or GTFO
Переглядів 1,9 тис.2 місяці тому
Oxide and Friends 9/23/2024 RTO or GTFO
Oxide and Friends 9/16/2024 -- Reflecting on Founder Mode
Переглядів 1,3 тис.3 місяці тому
Oxide and Friends 9/16/2024 Reflecting on Founder Mode
Oxide Compute Sled - Gimlet rear tour
Переглядів 1 тис.3 місяці тому
Oxide Compute Sled - Gimlet rear tour
Introduction to Oxide's Compute Sled - Gimlet
Переглядів 2,4 тис.3 місяці тому
Introduction to Oxide's Compute Sled - Gimlet
Oxide and Friends 8/24/2024 -- RFDs: The Backbone of Oxide
Переглядів 2,4 тис.3 місяці тому
Oxide and Friends 8/24/2024 RFDs: The Backbone of Oxide
Oxide and Friends 8/19/2024 -- Whither CockroachDB?
Переглядів 2,7 тис.3 місяці тому
Oxide and Friends 8/19/2024 Whither CockroachDB?
Oxide and Friends 8/12/2024 -- The Saga of Sagas
Переглядів 1,9 тис.4 місяці тому
Oxide and Friends 8/12/2024 The Saga of Sagas
I got the vibe that Paul Frazee does not love the idea of loving Mastodon...
This is tight
just awesome
I never imagined I'd still be watching people grapple with Zooko's Triangle in 2024. Tying accounts into DNS is a good start tbf.
I'd love to have been part of it in real-time (and not have missed the registration for the ticket/t-shirt), but I've listened to the VOD, which was very nice.
My apologies for the bad mic quality: Something must've gone wrong with my output settings.
Time table for each talk: Morning session: 08:42: Bryan Cantrill - State of the Union 1:08:10 Adam Leventhal & Ben Naecker - USDT: Past, Present and Future 2:11:15 Kyle Simpson - DError: Banishing Debug::fmt for nested enums 02:29:50 Alan Hanson - Dtrace: Building a Monitoring Tool 02:54:00 Aapo Alasuutari - UserSpaceProbeImplementation 03:16:35 Nahum Shalman - OpenTelemetry tracing for Dropshot Afternoon session: 04:34:30 David Pacheco - New user resources 04:51:33 dr. Kris Van Hees - Linux DTrace with BPF 05:33:46 Ryan Goodfellow - DTrace and P4 05:58:50 Robert Mustacchi - DTrace: Getting Elves to talk to Dwarves 06:15:54 Eliza Weisman - DTracing: Two great tastes... better together? 06:58:23 Luqman Aden - SDTs optimised into tail-calls panic
Timestamps: [00:08:42] - *State of the union* ( _Bryan Cantrill_ ) [01:08:07] - *USDT: **-Past, Present-** Present, Past, and Future* ( _Adam Leventhal & Ben Naecker_ ) [02:10:35] - *DError: Banishing Debug::fmt for nested enums* ( _Kyle Simpson_ ) [02:29:50] - *DTrace: Building a Monitoring Tool* ( _Alan Hanson_ ) [02:54:00] - *USDTs the wrong way* ( _Aapo Alasuutari_ ) [03:16:38] - *OpenTelemetry Tracing for Dropshot* ( _Nahum Shalman_ ) [04:34:35] - *New user resources* ( _David Pacheco_ ) [04:51:54] - *Linux DTrace with BPF* ( _dr. Kris Van Hees_ ) [05:33:49] - *"The delicious mashup between DTrace and P4" to quote Bryan* ( _Ryan Goodfellow_ ) [05:58:50] - *DTrace: Getting Elves to talk to Dwarves* ( _Robert Mustacchi_ ) [06:15:54] - *DTracing: Two great tastes... better together?* ( _Eliza Weisman_ ) [06:58:18] - *Diving into **_opte_**_#476_** "SDTs optimized into tail-calls panic"* ( _Luqman Aden_ ) God it was hard to timestamp this without getting sucked into each talk - awesome conf. everyone!
actual start at 8:42, setup etc before 😅
You're a real one
The 14nm to 10nm transition was the point at which everything fell apart. They fell off schedule and tried to paper over it by labeling their follow-on 2nd, 3rd gen 14nm at 14+ and 14++ like they were significantly different or better than the first gen 14nm. Meanwhile they could NOT get 10nm to yield and put it IN production. If Samsung and TSMC could get 10nm to yield and put it into production why couldn't Intel?
It seems obvious that Intel should be run by Ben & Jerry.
Intel has already build the microcode license manager, in 2010 and 2011 there where 4 SKUs where you could pay $50 to unlock "extra" cache and enable hyper-threading. Search Pentium G6951 if you want to know more.
Intel has always relied on Microsoft for the high-level roadmap. Microsoft missing the boat on mobile meant Intel missing the boat.
EDA software while being expensive is not the biggest part. Its the people , salaries and expertise. Building IP (CPU/GPU) is really expensive.
Totally fair -- and the world I would like to see is one in which the IP and tooling are both freely available to IFS customers. And sorry if any drink just came out your nose laughing uproariously at the 0% probability that that will ever happen... 😉
Lol, I work in SoC architecture and this was one of the best Oxide and friends episodes. I thoroughly enjoyed every moment. The cultural rot (not just Intel) needs to be studied more tbh.
Intel also screwed up their Altera acquisition. Also their fabs used their proprietary tooling, making things much harder for third parties to use Intel fabs.
YES. As it turns out, two hours isn't nearly long enough to talk about all of Intel's missed opportunities over the years, and the total and complete bungling of Altera was something I was kicking myself for neglecting to mention. Altera is especially painful because one could imagine some really amazing integration with a CPU provider if it were done the right way -- but of course, it wasn't. The best thing about the Altera acquisition may have been that it brought Xilinx to AMD, where I think better integration is at least possible...
by the transportation analogy, an ASIC would be a Formula 1 car: fast like no other, but it only really works for that one application
Oxide silicon when?
We should hurry and get in on that CHIPS act $$$!
Does anyone know what the pat gelsinger supercut is they talk about in the beginning?
Oh yes! ua-cam.com/video/RWHocP0AzMc/v-deo.htmlsi=3oK7CE_rBy6xEkUn
ua-cam.com/video/RWHocP0AzMc/v-deo.htmlsi=dyVioyfIdmMGfNsd
not my main takeaway but i read power of a pronoun for the first time because of this episode and . that's 2013!!! lot of respect for the values on display
I feel like blogging is a skill, one that needs to be honed and practiced regularly. I'm deeply jealous of Yegge, and I'm not sure I'll ever get close.
For the record, the image is discussed at 1:10:27. Adam did some sleuthing to discover the image, which is the New Yorker caption contest from 7/23/2007 (aside: I nailed the year!). Here were the finalists, also dug up by Adam: 1. "Have you considered writing this story in the third monkey rather than the first monkey?" 2. "We'll have to run it by our infinite number of editors." 3. "Sorry, guys. It's another rejection letter." I will let the Internet be the judge over those submissions versus by own, but, needless to say, I stand by my own sense of grievance from nearly two decades ago.
Don’t worry, folks, we’re getting him the help he needs ❤
@@ahl0003 It seems like this monkey joke has driven him bananas
I love all your blogs, this is the one subject as a non-hardware person I know you're qualified to talk about ❤
Thanks, I think?
I'm super excited about all of the innovation 0xide are doing here, and I was hoping that this conversation would answer the two questions that we're all thinking about: how much more efficient is rack-scale power delivery, and why is it more efficient?
This video is a companion to the blog linked in the description that answers those questions 🙂oxide.computer/blog/how-oxide-cuts-data-center-power-consumption-in-half
The recent blog post on the Oxide blog (mentioned in the video, linked in the description) answers those questions (I don't think this video makes a lot of sense without that context). A large part of it is easier to cool form factors for the single larger power supply, so a lot of the cooling gains discussed in the video are possible due indirectly to their power solution.
@@oxidecomputercompany Thank you, not sure how I missed that. So it's not that the power supply is necessarily more power-efficient itself, but that it's an enabler for more efficient airflow architecture?
@@capability-snob several factors, one of which is doing the power conversion once instead of many power supplies doing many conversions, in addition to the taller server form factor that allows for more efficient airflow.
@@oxidecomputercompany oh, "doing the power conversion once" = "not having to go from 3-phase to 240vac and then from there to the various internal voltages" ? I mean, you are still doing it twice, once to 60v and again to the various component voltages, but oxide don't need to worry about populating rails that don't matter, such as the overbuilt 12v rails many servers have in case you wanted to fill them with rust. I'm looking forward to the technical deep dive at some point!
Not long enough. Need more books
I sure in my whole life would not have expected Bryan to welcome me to the city where I live, but that just happened I guess. Grazie!
My goodreads stack will never recover from this. I already have 78 books in my oxide-book-club shelf from previous episodes.
Please check the date in the title of the episode :-)
oops!
Hello from Turin
Which episode is Bryan talking about at 1:39:05?
Thanks for asking! share.transistor.fm/s/65a10522
Beer Tunnel
We've spent the last 60 years - since Dennis and Van Horn - re-implementing capability security badly and re-discovering that these techniques happen to work shockingly well.
System 76 can make Coreboot work on modern processors.
So much respect for 0xide's communication around AI. If you have a use case, here's how easily you can do it, otherwise, there's not a shovel in sight.
I have to say, the sheer speed of provisioning and setting up Helix, really shows off how much Ooomf! is behind the Rack.
Sounds a bit like kql from Microsoft
UNIX is the outgrowth of that which exists in all of us.
If a SQL query is taking too long, then either the index is defined wrongly or there is no index, in either case the query is doing (a) full table(s) scan.
Is that "not invented here" I hear at 06:56? From anyone else, no surprise, but from you guys, the students of history?
Yep, you did reinvent the wheel, and not in a good way. Use ANSI SQL.
We’ll get on that!
Glad to hear it. UA-cam comments are by far the best source of software design wisdom. Always have been, always will be.
Piping tables reminds me of Power Query M code
M3QL was a pipe based query language for M3 TSDB eventually abandoned for PromQL.
I have to say, the audio has been consistently good last 4-6, 7 live streams. Levels have been good.
Oh interesting you're not compiling OxQL to SQL.
There is no on time, only early or late. Except for Gandalf, he arrives precisely when he intends to.
Oh God now I'M triggered. I'm a Deuce Deuce rider too. We have enough high-powered engineering talent in the Bay Area to turn this into a train paradise, if we wanted to! How do we make that happen? IDK
Step Nr.1 is to realize that trains don't operate in a vacuum. To make trains effizient, everything around the trains need to be reconsidered. First up, changing land use pattern and zoning. This is fundamental and the primary reason the US is fundamentally failing at doing great transit, even when spending lots of money. When your train station is surrounded by highways, stroads and parking lots, it can't succeed. Second, trains need a feeder system, walking, bikes and buses. Buses need to be have a coordinate time table with the train, clock face scheduling. In order to do this buses need to be given effective priority, that means removing car lanes and parking spots, in addition to signaling priority and fundamental road redesign around actual modern smart road engineering principles, rather then 60s fake-science. Then also redesigning roads to make them appropriate for real bicycle use as transportation. The problem is it doesn't require some new engineering, or brilliant startup, it requires no technology that wasn't already in use in the 1960s. Its purely a social and bureaucratic problem. But don't reinvent the wheel, just adopt Dutch road engineering and safety standard, Japannase zoning laws and Swiss train management practice. Don't get any Silicon Valley startups involved, simply go to a big European or Asian company and let them do it. You don't need more tech bros reinventing trains. So yeah, just change your whole society I guess. Good luck.
Some great takeaways from this! Was a good one to reflect on my own style of 'management' as I'm trying to navigate keeping a foot in the technical details while taking on more senior managerial type work at the startup I'm at. I think Clarity + Trust is a really great lens to use to decide how to get involved. I don't have the time or energy to micromanage, but I can sometimes be a little too hands-off. I've seen how people, particularly Engineers, can get caught on certain ideas and head down paths that don't really make practical sense to do. I'm not sure if you've covered it in a past episode, but I'd love to hear more about how Oxide manages to set the direction for individuals and how you make sure you're building the right thing, and delivering the right work, on a day to day basis. Like RFDs obviously seem fairly core to setting the direction, but how do you then make plans around that and deliver on it? At my startup we're basically mandated from up-high to be doing sprints and planning in Jira and all that. I try and keep it as low touch as possible while still giving the visibility that the execs need, but I find myself more and more saying "Can you make a ticket for that?" "Is there a ticket for that?" It's our only method for tracking the ridiculous amount of work we need to deliver on, without dropping it all. Also, I loved the detour into the elaborate prank on Dave Hitz
If I ever start a business, I'll probably be in Flounder Mode. Lying at the bottom, both eyes fixed in the same direction. 🐠
Was really hoping you would cover this. Thanks!
I don't know anythibg about servers. How do server CPUs work? Are they x86? Can they run CLI desktop programs the same way?
These are an AMD Milan CPU, x86-64 like many desktops/laptops, but with a lot more cores. I admit I don't know what operating systems they ship or support, but my guess is that customers will expect to run the same Linux and windows server workloads their off-the-shelf servers do. Yes, in theory anything that can run unattended is a good candidate for running on a server.