Extraordinary hardware in an ordinary system? The LinuxONE Rockhopper 4

[@KastanDay]

I was an intern at IBM Research and my manager was excited when I sent them a collection of your IBM videos. great stuff!

[@mycosys]

I think none of us have ever considered Z family because holy moley are they pricey unless ur actually one of the few people who needs high-reliability computing lol

[@TechTechPotato]

Hardware costs per system, sure, but I think the idea is that if you look at what you're replacing, what that cost, and licensing costs you save, it's a full TCO calculation

[@ChrisDupres]

​@@TechTechPotato so I spent a lot of years doing big enterprise consulting and the most common migration strategy from Z/OS mainframes (or to super high availability systems) is actually to IBMs Z Cloud. This is actually a really good middle step because doing mainframe programming is waaaaaay different than doing server programming. Took me like a year to wrap my head around connectors and how the different components even communicate to external systems.

[@nnasab]

I like the new server’s power consumption, that also effects the AC unit size for cooling too, so power saving is more than you thought.

[@MarkRose1337]

I appreciate this video. It's nice to learn about things beyond x86.

[@nisetsu]

Why you should NEVER consider IBM for anything? Remember how IBM has fired a bunch of older employees to screw them out of their pensions. That's just one good reason.

[@Brickoliciousness]

How does that work? I own my pension? Is this some kind of American only thing?

[@PeterDrage]

That must be a US thing as in The UK and EU the employee owns their pension and not the company.

[@IAmPattycakes]

​@@Brickoliciousness yes, in the US you can get fired and have your pension dissappear. So what IBM did was fire a bunch of people in their early 60s so that they didn't have to pay a penny of their pension.

[@kortaffel]

IBM had a scandal recently where they were discriminating older employees as "Dinobabies" and try to fire them by moving them around and stuff... Wasn't nice, I don't like discrimination of any kind

[@mehdimido5270]

That's not worst than Intel, AMD or Amazon. If you judge a company by how they treat their workers you won't buy a product from any fortune 500 company.

[@rakpiotr]

It is no secret that IBM did heavy lifting for linux multicore scaleability and many problems were solved there before we even could touch machines with that many cores. Things like RCU come to mind. Linux kernel support was never big question for me. After reading bit about OpenPower years ago, I wanted Power based workstation, to try out this platform. I just wanted experiment little bit with VSX, SMT4 among other stuff. Unfortunately the availablity and price made me to strongly reconsider. Something like this platform for sure is beyond my reach, if it doesn't pay for its bills and I don't even need to ask how good my TCO will be. For me, that means x86 will be getting most of developer mind share and they will struggle to build strong software ecosystem around Linux and FOSS. To add to that, in my former job (one of GPU manufactures) there was a some opensource library, that guys from IBM had to support on their own. Some HPC people were using it on Power machines somewhere. We did support both x86 and arm because that was what almost everyone was using. But sure, if that is what you need, you can deploy it and it'll work, this hardware looks really amazing.

[@velo1337]

would really like to see some backup of all those claims. especially on database performance and compute power. since it does not really make sense that the ibm cores are that much faster then x86 cores.

[@privacyvalued4134]

Same here. However, what you are looking for is probably something that the Linus Tech Tips channel is better suited to handle than TTP guy reading the PR lines from a teleprompter.

[@natedotson1397]

Just do some research on the IBM Z microprocessors and the architecture. They are custom designed and tuned for high throughput performance and massive scale for transactional workloads with massive amounts of on-chip cache that provide performance and latency benefits that are not matched anywhere. Architecture matters

[@asdkant]

In general, IBM Z tends to be designed for more enterprise-y, I/O-heavy workloads in mind, so for those it should work well. Then again, AMD is putting a lot of PCI-e lanes in their newer stuff so a comparison should be interesting. Although... there's a bunch of enterprise-oriented stuff in Z system besides the pure performance (LPARs, the redundancy stuff, etc.) that also make it into the equation for an actual customer, so do take that into account.

[@lawrencedoliveiro9104]

Mainframes were never designed for CPU speed. Indeed, the CPU is considered a scarce, expensive resource, so they would put a lot of smarts into the I/O controllers to avoid bothering the CPU with requests for service for as long as possible. The downside of this is, while they offer high I/O throughput, they cannot offer low interrupt latency. This means they are optimized for traditional batch-style mainframe workloads, not rapid interactive response.

[@christopherblare6414]

Very cool! I wasn't aware of this line. A little more than we can justify, but still neat. It is #sponsered and you even made a reminder joke/references in the content. I really appreciate that. Especially on a channel like this with mixed sponsored/non-sponsored content. That being said, I didn't like this video. It was a commercial. It felt like a commercial. I like hearing about unique and innovative hardware. But I come here for thoughtful comentary, not promotional fluff pieces.

[@PainterVierax]

What mostly feels like a commercial is the unique and brief allusion to Power. I mean why talking about x86 whereas OpenPower is already way more suited for this task and already scales to small racks servers? This point feels like IBM is just advertising their new stuff.

[@clehaxze]

I am able to get a Linux instance on the OSS community cloud provided by IBM (you need to be a maintainer and IBM approve your application). I believe it's running on the previous generation LinuxONE. The performance is amazing. 2 shared cores of that is crushing what I get from 2 Zen2 cores on my own dev box. I wish the price could be sane and I can own one personally. One of the standard rack mount ones will more then cover my development, hosting, homelab needs.

[@spuchoa]

Telum is a extraordinaire piece of silicon. Hope there is a comparison against Zen 4 using Phoronix test suite

[@CraftComputing]

Ian: Rolling cart with server out the door Me: I've seen that smile before 😀

[@peterjansen4826]

Linux on this channel? Miracles do still exist. ;p

[@keyboard_g]

Are people moving _to_ Z processors?

[@tulsatrash]

Exciting!

[@Fractal_32]

That is definitely intriguing hardware, I am a normal (Linux) desktop user so it would definitely be overkill for any application of mine but nevertheless I found it interesting. I’m always interested in server hardware and Linux given that linux offers customizability, speed and privacy due to it running free (libre) and open source software.

[@Shadowauratechno]

TCO seems great if you're running proprietary software at a large scale, but for homelab people running foss like me, it just doesn't make any sense. Although I suppose IBM knows that, given that they seem very opposed to selling to individual customers. It's a shame because I'd love to switch to Z or PPC64

[@shanent5793]

What does all that RAS cost in terms of performance? Are they useful for throughput workloads?

[@orthodoxNPC]

So do the cores have dual-issue pipelines? What makes them handle more with less? You skipped over a lot

[@reeldeelz2940]

Wait so these servers are not ARM or x86 .. are the apps recompiled to run on these Z architectures? or are they hyper-vising x86?

[@TechTechPotato]

Answered in the second half of the video. In short, both!

[@BillBroadley]

Fits in a rack and runs linux, sounds great. The only info I need is performance and cost. Sadly the "Cost, sustainability, and Software" section does not discuss how much the system costs.

[@TechTechPotato]

Cost is very usually customer specific. You engage with a sales rep and it's a full support package with hardware - no simple Lenovo dropdown here.

[@adibbins]

Is the CPU architecture supported by the big boy's like Oracle and SAP? We can assume IBM will support DB2?

[@Karthig1987]

Cool stuff

[@paulw3182]

Wow thanks for the video. The power delivery system is interesting but what is the supply voltage? In addition could you do a component deep dive in the future possibly focusing on a powered up system and the details of replacing various modules on the fly?

[@gesnow]

You can use I, p or z from IBM, power.

[@platin2148]

How is the uptime?

[@TechTechPotato]

seven nines. 99.99999%. 3 seconds a year

[@spewp]

AIX doesn't even get a look in anymore?

[@capability-snob]

I would totally move to power or whatever if our cloud provider supported it, as it's all just a recompile away. IO and Cache look pretty sweet. However, I don't get the point of the RAS stuff. If your system needs to be fault tolerant, you already need to write the software to handle an entire datacenter getting destroyed. If you need it to be correct, you already need to have written a validator process. So, I'm not sure who this stuff is really for. Thanks for taking a look at it though - quite a nice design.

[@henryzhang7873]

The number of companies that actually have real multi-datacenter fault tolerance is very small. Lots of companies (including some major apps that you probably use) can tolerate a DC failure in theory but very few systems are actually tested that way (I've worked on these, sometimes failover works but will fail during peak hours as well due to insufficient overprovisioning). The executives don't want to need to hire engineers who can program hardware fault tolerant apps because those people cost 2-3x more than the average engineer.

[@treyquattro]

you make advertising fun (it's a Fleetwood Mac song...)

[@Gastell0]

4:58 - 68 cores at 4.6GHz? Jeezebus, that's almost double that of AMD* at comparable core count

[@chaitanyakulkarni6416]

why volume is so low ?

[@adamreid5901]

Soulja boy telum

[@esra_erimez]

I just wish there was a z processor on an ATX motherboard

[@Steamrick]

Working at a medium-small IT service provider, none of our customers (small to medium companies, mostly under 1000 employees) have the budget or the need for it. Frankly, most of them would be fine with 90% uptime (so single nine), so long as the downtime is predictably far outside work hours.

[@DoctorX17]

Drooling... Man that's cool. Hopefully this means there's gonna be a flood of relatively new server hardware on the market for us poor folks, lol. Hopefully RAIMM will come to all of us one day

[@mycosys]

Hey Dr Cutress, can i request a video? The black magic of the Z family maintaining transactional coherency across multiple redundant failure tolerant drawers, first with synchronous processing, then some black magic to do it more efficiently asynchronously with the system controller in the later z family, and then....... Telum doesnt even have a system controller? What witchery is this?? (also think it would help some of your audience understand wtf the big iron Z family are)

[@christopherjackson2157]

Yes this is what Ive been waiting for since they first announced it. Haven't been compute constrained since epyc first came out, 100 percent io constrained Would love to sample one of these if anyone who's reading this is wondering....... Lol

[@AI-xi4jk]

There is always a question: how much?!

[@xerox445]

Soulja Boy Tellum

[@SirReptitious]

RAIN is a copyrighted b.s. term made by one of the SSD makers; I think it's Micron/Crucial. It stands for Redundant Array of Independent NAND and had nothing to do with memory. Don't get me wrong, I'm not having a go at Micron. I own two 2TB Micron 1100 SSDs and one 2TB Crucial MX500 SSD because they are great products that have been working 24/7 for many years now.

[@badcrc1]

in this context, it is RAIM. not RAIN

[@simivb]

I'm confused. Do you work for IBM now? The last time I watched this channel it was independent reporting, but now it just sounds like a marketing script. Pitty :(

[@TechTechPotato]

I'm an analyst, they're a client sponsoring videos. Intel, AMD, Qualcomm, and others are clients too. Says in the description. I've been busy with my day job to run regular content of late.

[@simivb]

​@TechTechPotato I guess my expectation about what sponsored content would look like was off then. I thought it would be more in the direction of you being paid for talking about something at all, but still saying your own opinion. This didn't feel like that. (No ill will!)

[@joebflies]

yea all his new stuff seems to be this infomercial type content its very sad. he seems to be cashing out on his past good content

[@terjeoseberg990]

Well, apparently it does break down. But it can be repaired without rebooting.

[@sambashton4966]

Watching this video it's like the last 20 years never happened. Basically every piece of modern software, from Kubernetes to MongoDB (two projects IBM are pushing!) are built to run as a distributed system which uses inexpensive individual nodes which can die at any time and yet system as a whole carries on. Why are IBM continuing with the hyper expensive keep the single server running at all costs approach when that paradigm is now undesirable?

[@krj15489]

Because if a disposable container running an ERP or banking transaction dies it means you might have lost millions of dollars. Also these systems tend to run databases measuring tens of TB that must have extremely fast throughput, 100% accuracy and ability to operate on the hardware without dropping a transaction. Distributed containers are great for heavy volume of web traffic but not for ERP.

[@sambashton4966]

@@krj15489 So why create this video pretending that MongoDB and Kubernetes workloads are a reason to buy these mega expensive servers if the real target market is people that want monolithic software architecture?

[@lawrencedoliveiro9104]

We have much more cost-effective ways of ensuring the necessary reliability nowadays. Google does it by installing half a million commodity servers--they cut costs to the bone on every part except the power supply. This is because their systems are robust enough to recover from lots of hardware failures, but the single biggest cost in running a hyperscale computing centre is the electricity bill. And Google can respond to your queries in milliseconds. Try doing that with COBOL code on an IBM mainframe.

[@galileo_rs]

In 2000 I grilled an IBM representative at a local Linux fest (anyone remember those?) With recent news about RH I guess we were all sceptical for a reason.

[@niro1960]

mine broke down 10min. ago...(flies away)

[@kai990]

after messing with IBM products and their horrible website for a customer project some years ago - i will never touch it again with a ten foot pole

[@ПётрБ-с2ц]

10:40 but there's no necessity in specialized hardware for quantum-protected cryptography

[@linuxgeex]

Only Ian can get paid to post a video using Z-Series to browse the web as a glowing PoC review lol

[@TechTechPotato]

aw yiss

[@jcugnoni]

And now Redhat is closing its source code.. Thanks IBM for trying again to lock your ecosystem and try the monopolistic route.. Good luck to convince the Linux community that you are a fair player now.

[@joebflies]

I used to love this channel but now everything is just a tech infomercial when he gets paid enough.

[@insu_na]

ha cool, that's my former home town.

[@GrishTech]

Why does rockhopper remind me of club penguin? Lol.

[@shocka007]

Go you halves on the missing trolley full of goodies

[@kellymoses8566]

IBM has amazing technology that they charge WAY too much for.

[@xPakrikx]

Locking to proprietary hardware ... "but it can run Linux !!!" ... nah thanks. OSS nad FOSH is future, like RISC-V

[@krj15489]

That's not generally a concern for the large businesses that purchase this gear. They need throughput and uptime.

[@PainterVierax]

@@krj15489 So why not going to the other high troughput modern platform from IBM, you know the Power one they specifically don't talk about in this video despite having an open source hardware foundation.

[@Prophes0r]

"It powers the smartphone you're probably watching this on." What? Why? Why would ANYONE want to watch videos on a phone? Phones are for making calls, texting, GPS navigation, taking the occasional "Right now" photo, and to get internet access while troubleshooting your router when it's down.

[@james2hackett870]

They just work mainframes, pretty much any bank outside of western infrastructure ie cloud, just use mainfranes, plus they have highest security level possible, and near 100% uptime with zero RPO at instruction microcode level

[@mousethefoo1230]

What they did not use Arch Linux, completely ridiculous Arch is the best. /S

[@Fractal_32]

What about Gentoo or LFS? XD

[@ronjatter]

Good content even if it does look like you filmed it on a 2000's webcam.

[@TechTechPotato]

GH5 and a prime lens, but yeah

[@ronjatter]

@@TechTechPotato You have to blame the camera operator then 🙂How was it holding that processor? Was it heavy? It looks like a work of art !

[@kellymoses8566]

When you buy IBM mainframe hardware you have to sign an agreement NOT to publish any benchmarks. That should tell you that the CPUs are very slow.

[@Moonblade042194]

Sponsored click bait video…, unsubbed