The plumbing layer as the new kernel

Posted May 3, 2012 15:55 UTC (Thu) by cdmiller (guest, #2813)
In reply to: The plumbing layer as the new kernel by paravoid
Parent article: The plumbing layer as the new kernel

The whole systemd, upstart, replace init push is pretty interesting to watch when taking a long view. In using linux since 92 including slackware, redhat, mandrake, mandriva, debian, and more recently ubuntu (just to name more popular names) along with a host of unix flavors on several architectures one notices various approaches for core items like the init system. Distros copied the best init tools from commercial unix (like chkconfig) and some form of sysv or bsd init scheme. Redhat and debian put in place secondary config layers (/etc/sysconfig, /etc/default, /lib/lsb etc.) while some special purpose distros used a bare bones init with some other script layer instead. Now we have the new options of systemd and upstart along with others (the article indirectly mentions openrc (why the indirect approach there by the way?)).

From my limited perspective the main problems new init schemes face is reconciling server needs vs desktop needs, and the enterprise desktop vs home desktop. Support for multiple hardware architectures is also important and should not be ignored.

Unlike the MS takeover of server roles from Novell/Artisoft/Banyan etc. by leveraging desktop dominance, the linux surge began from the server, supplanting unix application, database, web, compute, and file/print roles. Linux (via samba) currently competes with MS for file and print. Linux is starting to take over in virtual hosting. Linux has a lead on thin client desktops (fat, thin, vdi, multi head) and leads in the embedded space (may as well include android here).

The question for distributions should be, if a new plumbing layer or component thereof were to sacrifice continued linux dominance or growth in any of the above areas, could it result in reduction or reversal of linux desktop adoption? Careful consideration of plumbing decisions so as not to lock out alternative approaches should be a high priority.

to post comments

The plumbing layer as the new kernel

Posted May 3, 2012 20:55 UTC (Thu) by drag (guest, #31333) [Link] (3 responses)

I like the discussion, but I differ on a few points:

> Unlike the MS takeover of server roles from Novell/Artisoft/Banyan etc. by leveraging desktop dominance,

Actually a lot of people ran screaming and yelling from mid-size server systems because they were expensive to manage, purchase and use. Seems like most people that are into Linux are completely unaware of the IBM culture and the mentality it created in the minds of many people. Programmers of those were seen as huge assholes that did the things they wanted and used their specialized knowledge to hold the businesses that they served in effective bondage. That is if you fired your IT guy then you'd realize that nobody else could possibly manage the systems because he designed it so that only he could know how it worked. Then you had to hire him back as a consultant at twice the cost.

This made businesses very gun shy and distrust a lot of developers and their mentalities. Microsoft offered a escape from that. IBM was expensive. Novel was expensive. Unix was expensive. NT was cheap. Guess what people flocked to?

It's really not as clean cut as 'Leveraged the desktop'.

There are very good reasons besides that as to why Microsoft NT is now #1 server OS as well as the #1 desktop OS. Also it's helpful to erase all ideas of 'monopoly' when considering things.

> the linux surge began from the server, supplanting unix application, database, web, compute, and file/print roles.

Linux started as a desktop OS and was adopted by many people because it could run on cheap hardware and was compatible with applications coded for Unix. IBM, SGI, and friends put a huge amount of work into it and that is how we ended with Linux 2.6, which made it a very respectable replacement for Unix servers.

> Linux (via samba) currently competes with MS for file and print.

Not really. It's like saying that Linux competes with Microsoft for text editing. File and print for desktops is extremely mundane stuff and has been wrapped up as just a piece of much larger systems involving things like Active Directory, against which Linux is not competitive.

I like FreeIPA a lot, but doesn't deal with Windows which means it's a no-go in any desktop support situation, unless you are a running a extremely pro-Linux shop. Samba4 is still seems beta-ish, unfortunately.

> Linux is starting to take over in virtual hosting.

Linux _IS_ virtual hosting. Linux effectively _IS_ the cloud system. Linux is not utterly dominate in web hosting and web applications, but it's the most popular platform by far. Even Microsoft is using Linux virtual systems for their Skype infrastructure.

> Linux has a lead on thin client desktops (fat, thin, vdi, multi head)

It has no such thing. Windows is massively more popular in thin client setups now. Linux is popular host OS for VDI installations on top of Xen and other things, but the guest desktops and protocols people use on their thin clients are largely Windows.

X Windows and Linux can be made to work in certain situations, but they are not competitive. Citrix has a huge lead.

I have seen how people setup combination of VDI-based applications and thick client desktops to manage applications and handle security issues for large corporations. I have seen how people run VDI halfway around the world for call centers and other things. It simply couldn't be made to work using Linux and X. Maybe could pull a lot of it off with NX, but unless their proprietary of NomachineNX is very impressive it would be very difficult and expensive.

> and leads in the embedded space (may as well include android here).

I haven't paid attention to the situation for a while, but embedded systems are extremely diverse. Linux is more popular, generally, but Windows is still more popular depending on what type of system you are looking at.

Hopefully Andriod is making a big difference. Working with Sun in the past with Java (which is extremely popular for embedded systems) was very unpleasant and Oracle is shooting themselves in the foot with this lawsuit business, which is just ridiculous. Being 'java' with out technically being 'java' was a brilliant approach by Android.

The plumbing layer as the new kernel

Posted May 3, 2012 22:23 UTC (Thu) by cdmiller (guest, #2813) [Link] (2 responses)

> Novel was expensive. Unix was expensive. NT was cheap. Guess what people flocked to?

> It's really not as clean cut as 'Leveraged the desktop'.

First MS introduced Windows 3.1 workgroups and killed off Artisoft pretty quickly, as well as taking over smaller file server shops. File and print server vendors took another hit when Windows95 came along and killed the ability to easily run windows desktops and applications from the file server (one of their main features). The non MS file and print server vendors were left playing catch up trying to implement any changes to the dos networking stack MS felt like making, there were undocumented interrupts, file locking issues, etc.. I call that leveraging the desktop to gain server market share. Price differential was a factor, but there were plenty of technical hurdles put in place as well. MS leveraged their desktop dominance to win major portions of the server space.

> Linux started as a desktop OS ...
> that is how we ended with Linux 2.6, which made it a very respectable
> replacement for Unix servers.

Ah, I saw linux gaining most traction in the server space. Sure us unix fans could get a workstation via linux on commodity hardware, but that was a niche group of users while lamp servers broke out and dominated the web server market well before 2.6 existed, primarily displacing unix installations for web hosting up through and beyond the 2000 bubble. Much of the whole compute cluster market was overtaken as well, remember beowulf cluster mania?

> Not really. It's like saying that Linux competes with Microsoft for text
> editing. File and print for desktops is extremely mundane stuff

Say what you will but linux does compete in that space mainly through samba and cups. Yes one will still use ad etc. to support the windows desktops.

> It has no such thing. Windows is massively more popular in thin client
> setups now. Linux is popular host OS for VDI installations on top of Xen
> and other things, but the guest desktops and protocols people use on
> their thin clients are largely Windows.

What I see is ltsp and friends in the thin client space. Yes with spice on kvm etc. windows is provided via linux, other than that the proprietary vdi solutions which are windows only are very expensive, where many times the reason to go thin client is $ driven.

> X Windows and Linux can be made to work in certain situations, but they
> are not competitive. Citrix has a huge lead.

Most of the thin client I see is lan fat client remote boot, read only nfs or cow nbd root. Kiosk or edu computer lab setups. I haven't seen citrix for a long time unless mentioned as xen, but certainly wan access via nx, spice, citrix would be better than straight x.

> Windows is still more popular depending on what type of system you are
> looking at.

Windows embedded? Really? Closest thing I've seen in a while are some wince devices and obsolete windows phones. Heck my 2 yr old lg tv came with an offer for the source code. It runs linux. Beyond the phones many of the arm devices are linux based, linux really is gaining in the embedded space.

The plumbing layer as the new kernel

Posted May 4, 2012 13:14 UTC (Fri) by drag (guest, #31333) [Link] (1 responses)

> What I see is ltsp and friends in the thin client space. Yes with spice on kvm etc. windows is provided via linux, other than that the proprietary vdi solutions which are windows only are very expensive, where many times the reason to go thin client is $ driven.

It doesn't matter if they are very expensive. It works and people are willing to spend money on it. The initial cost of the VDI solution is often not going to be significant compared to other associated costs for deployment and management.

> Most of the thin client I see is lan fat client remote boot, read only nfs or cow nbd root. Kiosk or edu computer lab setups. I haven't seen citrix for a long time unless mentioned as xen, but certainly wan access via nx, spice, citrix would be better than straight x.

Well your personal experience is not inductive of what is happening at the world at large.

Ever seen advertisements on television or radio for things like "gotomypc" ? I am sure that more people use that service, alone, then all the entire LSTP or whatnot combined. It does stuff you simply cannot do easily with Linux. Nobody offers a comparible Linux solution.

Most people that would use VDI in corporations are not going to be aware of it. I really doubt that 95% of the people in my company right now realize that the vast majority of their applications are remote. The experience is completely seamless and integrates naturally into active directory and group policies.

Spice is nice, but it's not widely adopted and the Redhat management tools are unfortunately very expensive. Also Spice lacks a WAN profile, which makes it useful for LAN-only. I expect it will improve and it's already better then X Windows for remote desktops over LAN.

> Windows embedded? Really?

Absolutely.

> Closest thing I've seen in a while are some wince devices and obsolete windows phones.

You obviously don't know what Windows XP Embedded is. It is still very popular in POS, Koisks, "infotainment" devices and other such things.

> Heck my 2 yr old lg tv came with an offer for the source code. It runs linux. Beyond the phones many of the arm devices are linux based, linux really is gaining in the embedded space.

The embedded space is huge. Much larger then desktop or server space. Linux is extremely popular, but you have to be aware that there are lots of different segments.

The plumbing layer as the new kernel

Posted May 4, 2012 14:39 UTC (Fri) by cdmiller (guest, #2813) [Link]

> Ever seen advertisements on television or radio for things like
> "gotomypc" ? I am sure that more people use that service, alone, then
> all the entire LSTP or whatnot combined. It does stuff you simply cannot > do easily with Linux. Nobody offers a comparible Linux solution.

As gotomypc etc. are mainly used for remote desktop access and helpdesk support, they're irrelevant to ltsp deployments or vdi type solutions. Ltsp has worldwide adoption and usage in the spaces I mentioned, sure citrix and MS vdi / app streaming etc. are in use in big business sector. As for not being able to do that with linux, I would be surprised if it's not fairly easily doable but the market for it would be small.

> You obviously don't know what Windows XP Embedded is. It is still very
> popular in POS, Koisks, "infotainment" devices and other such things.

Fun story on xp embedded, colleague was walking down the street in vegas on his way to MS MMS conference, and one of the large bling signs on the strip had blue screened :)

And sure, embedded is huge, but classes of devices capable of running a linux kernel are growing. As mentioned in an earlier article, 50MHz arm boards the size of the old basic stamp are now in the $20 range.

Back to the primary point,

If a new plumbing layer or component thereof were to sacrifice continued linux dominance *or growth* in any of the above areas (mentioned in the original post), could it result in reduction or reversal of linux desktop adoption? Careful consideration of plumbing decisions so as not to lock out alternative approaches should be a high priority.

You appear to be arguing we don't need to care about linux participation in the areas we have been discussing? Is that your stance?

The plumbing layer as the new kernel

Posted May 4, 2012 4:36 UTC (Fri) by dlang (guest, #313) [Link]

the problem isn't that systemd introduces a new init system, it's that anyone who doesn't move to systemd is being painted as 'diverging' from the 'standard' or otherwise being 'divisive'

I happen to agree with Shuttleworth's comment that upstart knows what it wants to be while systemd seems to want to be everything, but the real issue is the conflect between people who seem to think that there should be multiple options vs the people who think there should only be one way.