Joyent Public Cloud EOL • Hacker News Board

Re: Joyent Public Cloud EOL

#161

Post by Annatar » Fri, Jun 07, 2019, 5:03 AM UTC

Thaxll wrote: post #91

Annatar wrote: post #24

Earlier quoted context omitted.

With technology which is an order of magnitude more concerned with reliability and easier to use than any competition out there, that's how.

By using Solaris / re-implementig KVM on different arch / platform, ofc it's the wrong approach, on one side you what 100s eng at Joyent working on that, where the rest of the world uses native KVM, x86, Linux ect ... it's just not possible on a tech level to compete.

SmartOS beats GNU/Linux feature for feature and it has features like fault management architecture that are science fiction for Linux. For example, it will migrate affected memory pages from portions of the chip with too many parity errors and it will off-line the chip if necessary, but the server will continue to function. All of that will be uniformly reported on top of that and is visible with fmadm faulty. I can off-line faulty processors and replace them on hardware which supports that while the system continues to serve.

Then there is COMSTAR, a reference fiberchannel and iSCSI implementation. illumos inherited the reference NFS V3 and V4 implementations from Solaris. The runtime linker is so advanced in respect to versioned interfaces that it looks like it's magical. The list could go on and on and on... Linux is nowhere near any of that. Linux barely works somehow but when it comes to running serious infrastructure and data integrity, that's not enough. Just because most people think bash completion and GNU commands are what makes GNU/Linux superior doesn't mean that it actually is: it's very crude and unreliable underneath, especially since you have armies of people constantly hacking on it and breaking kernel interfaces, especially in networking. Try running a firewall / VPN company with it to see what a nightmare it is. And that's just one instance.

And then there is cloud: nothing that GNU/Linux has comes even remotely close to how simple, secure and fast Solaris zones in SmartOS are. Literally nothing: I can pull an image with imgadm and instantiate a bare metal performance UNIX®️ or GNU/Linux virtual server in under 25 seconds with vmadm. No Kubernetes or Docker or OpenVZ or Terraform can do that, and those are all afterthought bolt-ons, while imgadm and vmadm are core of SmartOS and its reason of existence.

Re: Joyent Public Cloud EOL

#162

Post by Annatar » Fri, Jun 07, 2019, 5:19 AM UTC

chipotle_coyote wrote: post #114

Annatar wrote: post #46

Earlier quoted context omitted.

It's far, far from over as long as SmartOS remains easier to use and more reliable than GNU/Linux. The source code is there to fork, so just like GNU/Linux, there will be people working on it and using it. OpenBSD has 5.71 developers and yet they're kicking ass churning out security features and innovations like there's no tomorrow. I secured my copies of the illumos and SmartOS gates. Joyent is proof that large scal…

I'm sure SmartOS is very reliable once you get it up and running, but "easier to use" than Linux is one of the most "[citation needed]" comments I've seen on HN in a while. Perhaps for folks who've had long familiarity with Solaris, it's easy-peasy, but I never found an OpenSolaris distribution/derivative that was remotely friendly to Solaris newbies.

You have five commands you need to know: zpool, zfs, imgadm, vmadm and man. The rest are the same applications as on GNU/Linux. I don't know of any other system where five commands suffice to run it.

Re: Joyent Public Cloud EOL

#163

Post by craz8 » Fri, Jun 07, 2019, 5:24 AM UTC

rsync wrote: post #101

skrebbel wrote: post #28

Can anyone explain Joyent to me? My understanding always was that they're some special kind of hosting provider. But I never really got it. They sponsor Node.js, they sponsor all kinds of ex-Sun Solaris-related stuff, employing high profile people from both communities. What does supporting (and, at a time in the past, basically running) Node.js get a hosting provider? What about Joyent's products is related to Node…

"Can anyone explain Joyent to me? My understanding always was that they're some special kind of hosting provider. But I never really got it." Here is an old, historical thread that gives some context: https://news.ycombinator.com/item?id=4391669 I have a hazy recollection of their being part of the "free thumper" startup program that Sun Microsystems offered, briefly, and then later, blowing up one of those thumpers…

The lifetime hosting was clearly never going to work. Even when they launched it, anyone with any sense - and me - knew it.

But it got them cash and publicity, and didn’t kill them when they cancelled it, so, on balance, it made their business possible

Re: Joyent Public Cloud EOL

#164

Post by skrebbel » Fri, Jun 07, 2019, 5:36 AM UTC

bcantrill wrote: post #145

kragen wrote: post #135

Earlier quoted context omitted.

Solaris's death at the hands of Linux can be attributed in part not just to "the Sun attitude" but specifically to the attitude of Bryan Cantrill personally. In response to Dave Miller explaining how Linux's networking was technically superior (at that moment) due to more care for things like TLB footprints and system call overhead, and also had more accessible technical support, Cantrill famously decided to respond…

Ha ha -- wow! I had no idea that I possessed such power over time and space! As has already been pointed out to you (and has been pointed out repeatedly on HN over the years), I do regret the remark. So, there's that. I would also note that the rise of Linux in the late 1990s and early 2000s corresponds with the performance of x86 surpassing the performance of the RISC microprocessors during that same period -- a tim…

FWIW when I asked the "what's Joyent all about" question I never expected or hoped this would become a "let's blame Bryan Cantrill for random shit" flamefest. I'm not sure where all of that is coming from and I'm sorry if somehow I triggered all that.

You've been a great example to me and many others, in doing open source, in taking a principled stance (even if I subtly disagree with some of your principles) and also in showing that yes, it's totally possible to found a company with an incomprehensible business model, grow it, have it contribute lots and lots to awesome open source, and somehow despite all the free work and the incomprehensibilities, flip it for a profit. One day I hope to do the same and all of HN will be like "huh?"

Re: Joyent Public Cloud EOL

#165

Post by cryptonector » Fri, Jun 07, 2019, 5:50 AM UTC

Annatar wrote: post #14

60sec wrote: post #11

Earlier quoted context omitted.

Perhaps this has more to do with goals and focus than cost/efficiency. In many tech companies the scarcest resource is man-hours for development/maintenance. The order of complexity for public cloud/multi-tenant is much higher than for private/single-tenant, especially given the recent onslaught of CPU vulnerabilities. Still a big loss. I imagine it must feel like a defeat, especially to Bryan, given the corporate cu…

Yep, I could easily imagine Bryan being crushed by this. I know I would be, knowing I have superior technology to anyone else and yet being told I have to withdraw from the competition. It's like coming to a race track with a car and being disqualified because one's car is too light and powerful.

I don't know Bryan personally -I wish I did-, but I am pretty sure he's hard to crush. There's always something else to do.

Many of us Solaris Engineering refugees have gone on to do other things. And trust me, for me it was hard and still is because I don't yet get to use the new Linux eBPF stuff daily and w/o privilege for the equivalent of DTrace USDT, and I also don't get to use ZoL, and because systemd is a very poor not-SMF, and because there is no FMA for Linux, and because epoll makes me sad, and so on and on. But I don't dwell on all the mistakes made by Sun's executives from 2000-2010 (and they made so many terrible mistakes), not often. Since that seems popular in this thread, I'll tell you what I think those mistakes were (some of which Oracle is still making):

  - clinging to dying business (SPARC, J2ME) to the
    point that other, better business was forgone to
    avoid cannibalizing / competing with the dying
    business
    
  - killing Sun PS (2002)
  
  - insisting on unacceptable-to-Google licensing
    terms for Solaris (2002)
  
  - suspending Solaris x86 development (2002)
  
  - selling Sun to Oracle

There are more bad decisions, including buying MySQL (which was instantly forked, and that US$1bn price tag hurt).

Oddly, the torch of OS innovation has passed to MSFT. Not so odd, really, when you realize that MSFT learned the lessons of stack ranking (don't do it). And MSFT already had learned the lesson that to survive disruptions you must pivot. Bryan clearly learned those lessons early, so I think he'll be fine.

Re: Joyent Public Cloud EOL

#166

Post by hjhart » Fri, Jun 07, 2019, 6:05 AM UTC

sbarre wrote: post #90

joeblubaugh wrote: post #6

Wow, 5 months is an insanely short timetable to ask people to migrate entire systems to another vendor. I’m surprised their contracts don’t have better terms.

I would assume they've been discussing this move for longer with any customers who would need more time. Decisions like this are long in the making, this is just the public announcement for the rest of us.

As someone whose previous company has been on their cloud for five years, they didn't reach out to us until 3 weeks ago.

Re: Joyent Public Cloud EOL

#167

Post by hjhart » Fri, Jun 07, 2019, 6:08 AM UTC

jpgvm wrote: post #29

End of an era. Lots of good stuff to come out of Joyent and I hope we don't see the end of their OSS contributions. Manatee (the state machine behind Manta PostgreSQL HA), the SmartOS KVM port, Triton itself - huge body of great stuff in addition to Illumos contributions that they are perhaps better known for.

Manta was one of the most exciting technologies I touched five years ago. Color me impressed when I inline `sed` replaced 3TB of logs (2.5 years worth) in 30 seconds on our manta storage.

Re: Joyent Public Cloud EOL

#168

Post by SwellJoe » Fri, Jun 07, 2019, 6:10 AM UTC

pron wrote: post #131

SwellJoe wrote: post #102

Earlier quoted context omitted.

Technically speaking, they were quite far ahead of the curve. While VMs were all the rage, they were shipping container-based products (their Sun refugees kept developing Solaris Zones and ZFS in this context). The technical abilities of the Joyent folks were well beyond just about anybody else for a long time. One could argue they were too early. For several years they had the technology to beat Amazon, until Amazon…

I don't know if it was the immediate cause of their lack of success, but they had also brought with them from Sun the Sun attitude of doing stuff that their engineers thought was cool (and it probably was) without paying attention to how much people are willing to pay for it and whether their focus is correct from a market perspective. For example, Bryan Cantrill, Joyent's CTO, liked blaming Oracle's Larry Ellison fo…

"perhaps it was Cantrill who was monomaniacal all along, not keeping his eyes on the market"

Joyent made money and was acquired, all while producing metric shit tons of incredibly cool Open Source software. We should all be so "monomaniacal".

Re: Joyent Public Cloud EOL

#169

Post by bcantrill » Fri, Jun 07, 2019, 6:15 AM UTC

skrebbel wrote: post #164

bcantrill wrote: post #145

Earlier quoted context omitted.

Ha ha -- wow! I had no idea that I possessed such power over time and space! As has already been pointed out to you (and has been pointed out repeatedly on HN over the years), I do regret the remark. So, there's that. I would also note that the rise of Linux in the late 1990s and early 2000s corresponds with the performance of x86 surpassing the performance of the RISC microprocessors during that same period -- a tim…

FWIW when I asked the "what's Joyent all about" question I never expected or hoped this would become a "let's blame Bryan Cantrill for random shit" flamefest. I'm not sure where all of that is coming from and I'm sorry if somehow I triggered all that. You've been a great example to me and many others, in doing open source, in taking a principled stance (even if I subtly disagree with some of your principles) and also…

Hey, thanks for that! The HN comments here today have been full of surprises, not least the intense personalization of it all! I believe that our impact in the world is much more in the people we inspire than in those we rankle -- so in that regard, it's deeply gratifying to know that I have been of service; thank you!

Re: Joyent Public Cloud EOL

#170

Post by Annatar » Fri, Jun 07, 2019, 6:53 AM UTC

cryptonector wrote: post #165

Annatar wrote: post #14

Earlier quoted context omitted.

Yep, I could easily imagine Bryan being crushed by this. I know I would be, knowing I have superior technology to anyone else and yet being told I have to withdraw from the competition. It's like coming to a race track with a car and being disqualified because one's car is too light and powerful.

I don't know Bryan personally -I wish I did-, but I am pretty sure he's hard to crush. There's always something else to do. Many of us Solaris Engineering refugees have gone on to do other things. And trust me, for me it was hard and still is because I don't yet get to use the new Linux eBPF stuff daily and w/o privilege for the equivalent of DTrace USDT, and I also don't get to use ZoL, and because systemd is a very…

Apart from this entry I don't see any discussion on what Sun did wrong, but since you broached the subject:

- I'm furious about all the attention RISC V undeservedly gets while OpenSPARC is vastly superior in terms of mnemonics and virtual register windows and a free instruction every branch not taken and is GPL licensed;

- not devoting enough attention to i86pc Solaris was a major mistake;

- not selling cheap fully Solaris integrated desktops was the death knell: people knew Solaris from their student days, and universities had it because Sun made and sold workstations; once Sun decided to stop that, talent acquisition was stopped: desktop is needed for OS adoption;

- UltraSPARC T3 and T5 are blazingly fast, I know because I ran them, but now it seems too late;

- it wouldn't be too late if the RISC V nonsense pandering stopped and some OpenSPARC designed servers and workstations and even tinkertoys showed up by enthusiasts, like they did with Raspberry Pi;

- I'm glad Sun bought MySQL because Oracle subsequently destroyed it; MySQL is nowhere to be found now, and that is good so, because it was a shitty database which silently corrupts data; it deserved never to be made in the first place; that is one good deed in the world of damage which Oracle caused;

- killing OpenSolaris destroyed all that good will; had the project continued, it would have directly competed with GNU/Linux because we saw a constant monthly influx of GNU/Linux refugees on opensolaris.org forum, people who actually needed to get things done with their hardware;

- biggest mistake was that UltraSPARC based servers weren't cheaper than an average intel personal computer tin bucket server. That really killed them. And they would argue and refer me, a customer, to their internal "pricing committee" instead of listening.

What happens when a business doesn't listen to its customers?

"Bryan clearly learned those lessons early, so I think he'll be fine."

Indeed, and I wish him well and much success. He truly deserves by his deeds let alone his thoughts to continue being successful.