Yahoo and FreeBSD (1997)
zer0.org
zer0.org
I recently installed FreeBSD on a home server and was very pleased both with the process and the result. Also, the documentation is just impressive (that goes, in my experience, for all the BSD systems).
Note that Yahoo! was employing a few very bright freebsd kernel engineers but when the decision was announced that linux was the future many of them decided to leave. That made the path forward even clearer. It wasn't just a matter of integrating new acquisitions easier, and the fact that it was easier to hire people with linux experience than with freebsd experience. That was around 2008 IIRC. I left in 2010, and Rick less than a year later.
I don't completely agree with the reasoning, but it mainly came down to driver support, maintenance cost, and debugging tools.
Inktomi (YST) was acquired in 2003, later Overture (YSM) and both were linux shops. YST was the biggest property by far and we didn't have any kernel developers. We were doing just fine with stock debian, then stock RHEL. (I did a few very minor patches for 32-bit, like splitting user-space/kernel to 3.5/0.5G instead of 3/1G, and some caching improvements for /proc, but I wasn't a kernel developer). Linux just worked. Drivers were available and supported for the HW we needed. Oprofile, systemtap worked fine for the most part. We later hired one kernel developer to help Mail move to linux, but we always had more freebsd kernel developers.
For freebsd you needed to write the drivers and the tools. That took time and money. The community was also smaller and it was harder to hire people with experience. Acquisitions were running on linux too.
The decision was not easy. Y! already had great freebsd engineers. By announcing that linux would be the supported platform going forward it was a given that many of them wouldn't be happy and leave. Fortunately not all of them left. For example Peter W. still works there.
In any case I don't think there was really an alternative.
I think the justifications were better support for running on Linux (storage drivers, Java, MySQL, oracle), better support for virtualization (although bsd jails are better than virtualization in my opinion, and a better fit for Y!), and it would be easier to support one os instead of two and acquisitions (including inktomi) really wanted to run on Linux.
The drop dead date for getting off FreeBSD kept getting pushed back, but I'd guess they've moved almost everything by now.
That said, when a young upstart was trying to sell me on Linux superiority and described Docker to me, all I could do was snicker.
Of course it is.
Then there is a bug and you fix it just like any other security layer.
No it is not. Like Norton Anti-Virus, it has a much bigger user base. Other than that...
> Docker is available on FreeBSD (as of June this year) and uses jails.
Huh? I wonder why?
^sarcasm.
See https://www.youtube.com/watch?v=LrHW7Vvbie4 for a demo of what lxd can do. Maybe that will clear it up for you.
lxd has absolutely no relevance to this discussion. They are not jails, they are not like jails. As noted in my OP in this thread, jails are not a replacement for hypervisors.
> They are not jails, they are not like jails.
That's what I've been saying.
> That's what I've been saying.
No, you've been saying Docker is so much better than jails. Either you don't know jails or docker or both, I can't tell. Just you really want Docker to be awesome.
As I was my own jailer, lack of resource limits wasn't a real problem. I greatly benefited from the lack of performance penalties; everything ran at the same speed in or out of the jail. I had one or two systems running in VMs and they were terrible: a) things were clearly slower, b) virtual disk provisioning is terrible, c) timekeeping (hopefully things have gotten better since 2011)
Back in the day, Oracle did a freebsd driver partially for Yahoo. The Oracle drivers were not open source and hence there were few options to connect without Oracle's help.
I worked in the kernel networking group at SGI in the 90s and there was a lot of freebsd loving there. Also, the FreeBSD license was more relaxed and commercial products (like NetAPP) could include and extend FreeBSD without disclosing their modifications.
Our frustration with lack of support for FreeBSD moved us to choose Linux and Windows (for SQL server support) the next time around in 2004.
It's my impression that the pace of FBSD development has been increasing in the last couple of years, hardware/driver compatibility included. As well, it seems recently there's been more "cross fertilization" among the BSDs, and Linux to a lesser extent, signs of health and vigor for these projects.
While there are former FBSD users deciding to go with a different OS, such switching is not new or a one-way thing. I think it's a safe bet some enterprises are choosing FBSD over something else, anyway there's little risk any major FOSS OS project will soon disappear.
I think if FreeBSD would have put a bit more effort into modernization and community in the early to mid 90s we'd all be using it more now. It's guts were superior at the time and in a few ways still are.
And in particular, right now I'm grateful that the BSDs are small enough to be ignored by the herds of people migrating away from Windows and bringing their mindsets with them - that all seems to be landing on Linux.
RH for instance seems to be heavily pitching RHEL (never mind Fedora and CentOS) towards corporate and the Military-Industrial complex. The latter in particular are interested in RHEL after some poor experience using Windows on warships and similar.
But at the same time there is a pile of MSCEs running around in these organizations, and so there is a incentive to make daily Linux management at least superficially similar to reduce retraining costs.
Returned it for The Complete FreeBSD and never, ever looked back. FreeBSD got me jobs at interesting places, and a handful of life long, interesting friends.
I agree with some of the other replies here -- I'm perfectly happy and even glad the BSDs have exactly the size and composition of users they do.
well, NetBSD needs much more public love, in my opinion.
What made me move to Linux eventually, was the ability to keep all my software easily updated, including the kernel.
Honestly, I don't know if this is still the case.
It makes it difficult to support the applications and services that run on the OS as a result.
Whereas with something like RHEL/CentOS, we're guaranteed stability for years and can patch without too much concern of major breakage.
This is the part that really appeals to most enterprise users.
What _is_ important is ABI stability of the kernel, because it lets you figure out how to deal or not deal with Long Term Suck on your own terms and lets proprietary software vendors figure out independently how they want to do deal with it. Both Linux and FreeBSD kernels do this pretty well.
I thought it was a nice idea, and they really push this as being "an easy thing anybody should be able to do".
which fits with my philosophy of "severs should not have/use compilers"
If I can get a shell on your server I will just download my own compiler.
I don't need source files, I need a machine to be a compute node.. or a storage node.. what use is a compiler if I do it right..
the new generation of this is: "I should never need to shell into my machine if I did config management right."
You can use `freebsd-update` to update the OS itself from binaries instead of having to build world. https://www.freebsd.org/doc/en/books/handbook/updating-upgra...
You can use also `pkg` to install/upgrade software packages like `apt-get` on Debian derived distros. https://www.freebsd.org/doc/handbook/pkgng-intro.html
freebsd-update.conf
> Both freebsd-update and pkg are no way close to apt-get and more importantly Debian 'unattended upgrades
Exactly, it's better.
Any how, aside from an NT server or two, and a Sun IPX box, our servers were generally Linux (Slackware and Debian) and FreeBSD. Mostly FreeBSD.
FreeBSD had a lot going for it. A lot going for it over Linux at the time, for our company. Often we would get our hands on an x86 box of one type or another and have to make it into a server. These boxes usually had no CDROM, and usually we installed network cards, which themselves were coming in randomly, on their bus (usually ISA, sometimes PCI).
To get the latest version of Slackware, we'd have to download all five of the Slackware base "A" disks. Then the compiler was on the 10 "D" series of disks. Then the 4 "N" disks would have networking. Plus more if you want ghostscript (another set) or emacs (another set), or God help you, a workstation with X-windows running fvwm...
FreeBSD was two disks - boot and root. Then you go into ports and install from there. Because boot and root were enough to get your network card working - when Slackware and Debian often did not have those cards working at all. Many a time I was going to install Linux, couldn't get the network card working, put in the FreeBSD boot/root disks, got the network card working with no problem, and the would-be Linux box became a FreeBSD box.
It was not just network cards - these two disks had a great system for not just recognizing all Ethernet cards and getting them to work, but being able to install a full-out system over a modem and 56k (or 28.8k, or 14k) baud POTS connection. FreeBSD just made it real easy to install itself.
This was not just my opinion, others I knew thought the same thing. Not until Ubuntu did I find a Linux system that worked to make installing it easy. When wireless ethernet cards became more ubiquitous, I had left FreeBSD behind years before, but most Linuxes had trouble with every other card out there, even Ubuntu initially. FreeBSD never seemed to have these problems, being able to install easily was something they prioritized - and I think it helped them.
Also, our Linuxes at the time were vulnerable to the "ping of death" and these sorts of security problems. I still know companies which use FreeBSD. They started years ago and never changed.
I'm not sure when and why Linux started overtaking FreeBSD. Linux had this pre-GRUB boot loader called LILO which was horrible, but it did allow people with Windows boxes to be able to turn their desktop or laptop into a dual boot machine on which they could play around with Linux. Linux was more advanced on the multi-boot front if I recall, and that was probably one of the reasons it got ahead.
Also another mentioned reason is commercialized BSD like BSDI was under a cloud of lawsuits in the early 1990s, whereas commercial Linux companies like Red Hat were free of all of this, and this served to help Linux as well.
I believe it was almost entirely due to the work of Donald Becker. I don't know anything about the guy, other than he had a NASA email address, but I kept seeing his name mentioned when looking for anything network card related.
I remember the floppy disk pain though. Few machines had CD-ROMs so basically it was copying 30 images overwriting Microsoft Windows & Office floppy disks (they were better quality than the cheap 3.5" disks you generally got supplied with and there were always spare ones) and then swap, swap, swap...
Or maybe BSD has maintained its technical superiority in some areas for some users.