Slashdot Mirror
After 2 Years of Development, LTSP 5.2 Is Out
The Linux Terminal Server Project has for years been simplifying the task of time-sharing a Linux system by means of X terminals (including repurposed low-end PCs). Now, stgraber writes "After almost two years or work and 994 commits later made by only 14 contributors, the LTSP team is proud to announce that the Linux Terminal Server Project released LTSP 5.2 on Wednesday the 17th of February. As the LTSP team wanted this release to be some kind of a reference point in LTSP's history, LDM (LTSP Display Manager) 2.1 and LTSPfs 0.6 were released on the same day. Packages for LTSP 5.2, LDM 2.1 and LTSPfs 0.6 are already in Ubuntu Lucid and a backport for Karmic is available. For other distributions, packages should be available very soon. And the upstream code is, as always, available on Launchpad."
With 14 contributors, that they got it done in two years is impressive. Hopefully with this update, more distributions will be able to readily support LTSP 5.2 again...
This is great news to hear. I've been using LTSP at a school for all teachers and students since 2003 with excellent results.
Yes, LTSP has been around for more than 10 years now and is really about making all that easier. It lets the user choose if he wants to connect to a Windows server using RDP or to a Linux box either using X11 over SSH or just using ssh for authentication and X11 clear on the network for better performance. The main addition is having a login manager for that which can call a lot of hooks, mount the home directory directly on the thin client and then mix local and remote applications.
LTSP isn't some X11 extension. It's a distro that sets up a server, and supports a large number of diskless thin clients, with lots of features like running apps locally, accessing local disks, etc., etc.
Running X11 apps over SSH has NO RELATION TO WHAT LTSP DOES.
Slashdot gets worse every day... Pipedot: News for nerds, without the corporate slant
Yeah, you can do it without LTSP. But you don't have to be a guru to stand up an LTSP server and host desktops for thin clients so it's handy for the schools who use it. I've been using it at home for years to host desktops for guests because when the nieces and nephews come over they have incredible computer corrupting skills and need a platform that's less amenable to viruses than my kids' desktops and laptops.
You can also mangle the config to merge in DRBL, which allows me to netboot compute cluster nodes that I get at surplus if I want to do a little recreational number crunching or transcoding. I think it's pretty cool that we live in an age when an ordinary elementary school can have its own supercomputer and if their networking is up to snuff, join the ranks of the world's most powerful supercomputers.
But go ahead and rain on their parade if it makes you feel 1337.
Help stamp out iliturcy.
Maybe they check their work before they commit it. ;)
On the serious side, I doubt most of them were working on this full time. You may look at it more like this: every 10 days, they had time to actually commit something they were working on while working on other [I presume?] full time jobs?
LTSP isn't some X11 extension. It's a distro that sets up a server,
Sigh. LTSP is a system mostly utilizing existing servers that provides XDMCP over SSH, and the rest of your comment is true. But it's not a Linux distribution; it's software for Linux. And while it does have a lightweight login server, the really interesting parts are the ability to use local machine resources (as you say, disks and other USB devices) and that it sets up the netboot configuration[s] for you.
5.2 is a bugfix release for 5.199 and as compared to 5.0 most of the improvements are in speed and so-called 'fat client' support, where most apps are actually run on the client. There have also been tweaks to the way audio is handled.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
It is if they're also teaching school. A lot of open source projects are funded by enterprises that pay programmers to improve the project and contribute back. There isn't a lot of call for that in LTSP yet - corporations find it easier to just license Citrix or Microsoft Terminal Services.
This may change as VDI initiatives take off for the power savings, security benefits and management economy. Terminal services on low-watt thin clients is hugely green. LTSP for terminal services, virtual desktops in KVM, plus dedicated servers for high-powered users makes a nice blended VDI solution that suits low-needs customer service people who only need a browser, all the way up to engineers - as long as everybody can use Linux. Management overhead is very low and security can be as high as you can get with network access terminals.
Help stamp out iliturcy.
LTSP is supported by Debian, Ubuntu, Fedora/RedHat, Suse.
On another level, another important effort has been done to provide enterprise grade thin client (LTSP-Cluster) and all those efforts happens outside the LTSP tree and are not accounted for in this changelog.
After I started using NX ( http://freenx.berlios.de/ and http://nomachine.com/ ) I never looked back at VNC. NX provides near local speeds for remote X11 desktops.
But do you have sound working? Can you plug in a usb drive on your client and see it on your desktop? Yes you can do all that, but you also need to manage it, and LTSP supports installations of as large as 30,000 desktop clients.
The point is to have a thin client work like a desktop.When it works it is a huge time saver for admins, as you have one server cluster to back up and maintain, and a bunch of diskless workstations that can be unplugged and replaced with no configuration or installation beyond maybe needing to put the mac address in dhcpd.conf
LTSP is probably more work than it's worth if you have less than five work stations, but more than five workstations, and the long term savings probably make it worth the initial time investment, as you need beefy servers and good redundancies, but after you have a high availability server cluster that people are logging into, management as become a lot easier.
Work bio at MMWD
After almost two years or work and 994 commits later made by only 14 contributors, the LTSP team is proud to announce that the Linux Terminal Server Project project released LTSP 5.2 on Wednesday the 17th of February.
That's about one commit per 10 days per person. Is this sort of number normal in the open source scene? It seems very low to me.
They are probably using something old fashioned like CVS where all commits are globally visible and nobody can commit anything which might possibly break a branch. In mercurial I tend to commit every time I make a change and then collapse commits into logical patches before I push upstream.
http://michaelsmith.id.au
That's about one commit per 10 days per person. Is this sort of number normal in the open source scene? It seems very low to me.
It depends on the size of the average commit. If most of them are small changes to single files then yes this is probably slow. But if a developer is working on a complex change then an individual commit could represent a significant number of man-hours developing, unit testing and regression testing before the commit. This is especially true if they are using some form of distributed source control whereby said developer has a local repository to keep inter-commit changes tracked in or if they are using a branch+merge arrangement (so the developer commits partial changes to a "personal" dev branch) and they are only counting commits/merges to the main branch/trunck in the above count.
A browser and a VT-100 terminal are all that a lot of customer service people need and should have. The limitation of using a web application prevents a lot of activity you don't want customer service people doing like installing applications, running scripts embedded in documents, etc. Web interfaces have come a long way.
Likewise networking and thin clients have come a long way since the days of Token Ring, which peaked at 100mbps in the late 1990s. Thin clients have gigabit network connections now and every port is switched rather than being part of a bus or loop.
Most especially servers have come a long way. It's not unusual to have a 1U server that runs 16 3GHz threads on 8 cores, or 12 threads on 12 cores, using high-bandwidth/high IOPs SAN or local storage and 10Gbps networking. Back then 1GHz was fast for a server. 1GB was a lot of RAM, and today 192GB is easily reachable. Next month we get the 12-core 2, 4 and 8 socket boxes for up to 96 cores per server. This is just the commodity stuff - I'm not citing the special purpose stuff like Sun and Itanic for the obvious reasons. Heck, these days the SSD hard drive in my laptop can do over 8K IOPs - I can configure a server to do well over a million. Storage infrastructure also enjoys the leverage of newer technologies that leverage abstraction in new ways. You can, for example, create "smart clones" of a desktop virtual machine which work as deltas off of a "standard image" and require almost no storage at all. As the user uses it, the smartclone image file on the SAN grows only as much as the data written. As soon as the customer logs out, their temporary data is erased and no storage is consumed - and they get a fresh image the next time they log in which improves security immensely.
So in short, time sharing was bad back then because you were sharing from a very shallow pool of resources through a thin straw. Now the pool is deeper enough, the straw is wide enough, to give the benefits we were promised back then and didn't see. The clients, the network and the servers all have the capacity to deliver an outstanding experience. Sharing is an even better idea now because the drives, servers and even individual processors or cores can power themselves down and up based on demand and keep a reasonable amount of resources available to handle demand spikes.
The question now becomes whether or not we can return to the cathedral - the ivory tower of precious resources husbanded and defended by a heirarchical information clergy steeped in knowledge and cloaked in the mysteries of keeping it running and making it safe. We needed the Bazaar to improve productivity when the infrastructure wasn't up to snuff, but it's proven a costly and vulnerable environment for business. Getting the end users to give up their local autonomy is not going to be a soft sell - it's going to be a long and ugly fight. IT pros can probably ease the transition by making the virtual or shared environment more open and faster than the local one until the transition is complete, and then shutting down the ability of end users to do unsafe things once the migration is complete.
Help stamp out iliturcy.
Hello there ye from the realm of outdated misconceptions. Who told you terminal servers were slow? A few years ago, I set up a 20-seat Linux terminal server network. One terminal server (which was about as powerful as a high-end desktop machine at the time) was enough to handle all of the clients simultaneously and not break a sweat. We're talking P2-era technology on the client side. Each desktop was just as responsive as if the input devices were connected directly to the server itself. I talked to more than one person who absolutely could not believe that the applications they were using were actually being run on another machine over the network.
Strawman argument, nobody said either of those things. Even if they were true, the point is moot. LTSP-enabled distributions provide complete Linux desktop environments. You get sound, access to local devices, and all of the applications that run on Linux. About the only applications I wouldn't recommend for terminal servers would be those with demanding video requirements like 3D games, CAD, and video editing.
In a previous job, our whole office ran off a terminal server and it worked great for years. I did system administration and web development, never had a single task that the setup couldn't handle. We even ran dual-screen on the thin clients and never had a problem.
Be sure you count yourself in that 95%.
You don't do exactly what the project says it does. The project uses diskless system with no ROM OS to boot to a full X session or RDP session. Jim McQuillan basically wrote a system to duplicte what he was doing for hospitals at the time. When I used LTSP often (from 0.9 to 4.0, circa 2000-2004), the process worked like this:
I'd also like to give a big shout out to Eric Harrison, who made the whole system easy to use for schools with K12LTSP (now K12Linux).
Put identity in the browser.
LTSP is an outstanding product that scales incredibly well compared to other virtual desktop solutions. While a little off topic, LTSP Cluster is an excellent addition to large scale LTSP deployments. https://www.ltsp-cluster.org/
We ran 130 Xterminals (Linux machine thin clients) over switched 10-base-T with a 100-FL backbone for many, many years (up until 2 years ago). It worked just fine. The only thing that will kill the network is trying to play video or have Flash, neither of which we support.
Now we have 160 over switched 100-TX with 1000 fiber backbone. It is faster, but not THAT noticeable.
Sorry, but it's you that doesn't know what he's talking about, since you're too "time share" phobic to find out what's really going on.
Give me an enterprise-class machine (disk access, etc., though two machines would be better) of he same caliber as a high-end desktop from today, and I can run 15-20 cheap diskless clients off of it in a Gbit, switched environment, with the same performance of a low-end desktop from today. 3D accelleration. All the desktop apps. Local storage and printer access. How does that work? The performance comes from shared application libraries among different clients and cached memory on the server.
Yeah, I've done this in production environment. I've done it in my business. I'm not a jobless troll. It works. It saves a ton of money. Users have no idea anything's different than a standard desktop, except that booting is five times faster, and when hardware fails, I can rip a client out, put a new one in, and have the client back up and running in 5 minutes. I fix the broken client on my schedule instead of holding up work for it.
Stop being afraid and learn some new tech.
Put identity in the browser.
You can use LTSP to boot directly from bare metal to an RDP login. Does that count?
Put identity in the browser.
Can you even buy hubs any more?
I believe that all of those cheap 10/100 switches are really two hubs with a switch connecting the 10mb to the 100mb. Technically, there is a switch in it so they call it a switch - but it acts just like a hub.
And that's why we implemented localapps. Running firefox as a localapp will let you do fullscreen flash just fine. As I mentioned, LTSP is either using X11 over SSH or SSH only for authentication. In the second mode, your credentials are sent securely but the actual X11 events are send unencrypted, so that's actually faster than any OpenVPN/IPSEC you may use.
Who told you terminal servers were slow?
Experience.
It all depends on what is being "served" by the TS. However, I've seen huge servers brought to their knees by processes such as Powerpoint (or OpenOffice clone), or rendering images, or whatever sucking as much CPU as possible to run. Load that up by three or four people and now you have a TS brought to its knees. A server designed to host 30 to 50 clients suddenly can't support 10, all because people didn't include "Powerpoint" in the spec, because nobody heard of it when it was implemented, and a year later, someone hear of it, and now you have computers that don't function well for ANYONE.
If someone goes down the whole Terminal Server route, they best be understanding the dynamics of how people use computers changes and evolves, almost always to use MORE computing power, rather than less.
And to compensate for these changes, one ought to budget into the specification the upgrades needed to keep up with the increasing demands. Either that or start telling people "NO" when they want Powerpoint. And good luck with that.
This is not to say I'm against Terminal Servers, because I'm not. In places that have a limited demand for applications, a TS is probably an awesome solution for managing workstations.
Just saying you haven't experience it being "too slow" doesn't mean it doesn't happen. I've seen it happen, and it sucks when people expect a system to work as designed when the parameters of the system has changed.
Agent K: A *person* is smart. People are dumb, stupid, panicky animals, and you know it.