Facebook needs someone to make Linux as good as FreeBSD.

DutchDaemon

DutchDaemon

Administrator
Staff member
Administrator
Moderator
Developer
Good luck, Mark! Having fun looking at WhatsApp's backend?
 

Attachments

  • Screenshot_2014-08-07-00-03-51-1.png
    Screenshot_2014-08-07-00-03-51-1.png
    135 KB · Views: 1,802
I can appreciate them wanting to try and fix these problems in Linux rather than switch to FreeBSD. They probably have a huge investment in software for Linux at this point (deployment systems, monitoring, custom kernel tweaks etc) and it's easier to improve networking than switch the whole system to another OS.

Makes you wonder how many companies there are out there though that might have ended up with better products, or more consistent, manageable services if they'd gone with FreeBSD in the first place rather than jumping on the Linux bandwagon. Companies like Netflix and WhatsApp seem to be perfectly happy with their choice to use FreeBSD (there's a great talk by one of the Netflix engineers at a recent BSD conference on YouTube). Linux may have more active desktop support (where I use Windows or OS X anyway) but I really can't stand the thought of running services on something that's so unorganised and changes direction so often.

I guess it's also a similar benefit to the one that gets Windows so much Internet server use. If you can pull people in on the desktop, you not only build a large number of people proficient and comfortable in using the OS, but they are also much more likely to use that OS to develop/deploy server applications. The Facebook guys were probably already running Linux on their laptops so it was the obvious choice.
 
Note that they think they need a couple of years to even match FreeBSD's network proficiency. With a lot of staff. And, of course, it is a fast moving target.
 
DutchDaemon said:
Note that they think they need a couple of years to even match FreeBSD's network proficiency. With a lot of staff.
Click to expand...
Not impossible to do of course. Facebook does have very, very deep pockets and can easily afford highly trained engineers. I think the biggest question is, would those changes be allowed to be added to the Linux kernel? They can change whatever they want but Linus (and others) may not agree with their modifications. I think that would be their greatest hurdle.
 
SirDice said:
So, why don't they just use FreeBSD instead?
Click to expand...

I would think the answer would be painfully obvious. Facebook has thousands, possibly millions, of servers running Linux. These servers have been working well for Facebook for years. Hacking together a slightly better network stack for Linux will be much much less expensive than replacing all those servers with FreeBSD. Plus, FreeBSD would have to be customized and all the Facebook-specific tools would have to be ported. So the Facebook team can either make some adjustments to one small aspect of Linux to get what they want, or they could completely replace their massive infrastructure and then work on all the issues FreeBSD will bring to the table.

I'm not saying FreeBSD is bad or needs fixing, I'm pointing out Facebook is a highly customized environment and one cannot simply drop in a replacement. Any OS would have to be tweaked and customized in order to replace what Facebook already has in place. Then there would be testing and retraining for all the admins... In short, they can spend around a million dollars improving Linux or they can spend hundreds of millions switching to FreeBSD, the solution here is obvious.

Also, I would like to point out the title of this thread is wrong. Facebook does not want its Linux network stack to be "as good as FreeBSD". The ad clearly says they want to make the Linux network stack better than FreeBSD's. Obviously you cannot get that effect by simply adopting the technology you want to be better than. Since they want something better than FreeBSD they would need to re-write the networking stack regardless of which OS they were running.
 
The ad says "rival or exceed". So, at least as good as FreeBSD's. And yes, of course the title is facetious.
 
It's still nice to see an extremely Linux centric company talk about the superiority of the FreeBSD network stack. It still surprises me sometimes how many new startup companies go with Linux given the corporate friendly nature of FreeBSD. Not that there's anything wrong with that per se, but there are a few places where FreeBSD just makes more sense.

--
@@fonz - Doh! Thanks :r
 
Last edited by a moderator:
protocelt said:
It still surprises me sometimes how many new startup companies go with Linux given the corporate friendly nature of FreeBSD. Not that there's anything wrong with that per se, but there are a few places where FreeBSD just makes more sense.
Click to expand...

I suspect a lot of companies start off with Linux, not necessarily because it is the best choice long term, but it is something that is easy to set up and get started with. Most of the mainstream Linux distributions are pretty easy to get up and running in just a few minutes. Once a company is running Linux, is using tools built for Linux and has everything configured to use Linux it doesn't make sense for them to switch. I suspect more shops would get started with FreeBSD if there was a FreeBSD equivalent of SUSE's YaST or Zentyal. Those tools make even complex setups a quick point-n-click experience with little overhead. For most small companies the benefits FreeBSD can offer don't tip the scales.

TrueOS or ZFSguru are probably as close to those easy-access solutions as FreeBSD gets and even those have a ways to go. FreeBSD is a great operating system, but lacks a lot of the utilities and ease of use admins at small companies are looking for.
 
NewGuy said:
I suspect a lot of companies start off with Linux, not necessarily because it is the best choice long term, but it is something that is easy to set up and get started with. Most of the mainstream Linux distributions are pretty easy to get up and running in just a few minutes.
Click to expand...
I think you are totally wrong. As somebody who has over 50 Red Hat servers and desktops under my control I find painfully difficult to do most things comparing to OpenBSD (used for network infrastructure) and FreeBSD/DragonFly (file servers). Any serious company have people with enough technical competence to set up any operating system. There are two principle reasons for the success of Linux at least in U.S.

1. There is a legal entity you can sue (Red Hat, Novel, Canonical come to mind)
2. Vendor support. FreeMat, Scilab and GNU Ocatave are nice toys but unless MATLAB runs on FreeBSD my computing nodes will run Red Hat. The same goes with Oracle Java and many other common things.

Finally people are truly delusional if they think that the quality of the technology is what matters on the long run. If that was true most of use will run Big Indian DEC Alpha-s with some kind a O.S. which is built from lessons learnt on UNIX and younger counterpart VMS but this is really not the case.
 
I disagree.

1) A lot of people used to say that's the reason they run Microsoft software and Windows yet does anyone know any company that ever sued Microsoft for these reasons?

2) Not vendor support but software availability. However only companies that use that type of software care about those things and I wonder, relatively, how many don't. I've never worked for a company that uses MATLAB or Oracle anything.
 
pkubaj said:
What software do you need that isn't available on FreeBSD? For MATLAB there is math/matlab-installer, there are also ports for Oracle Java in /usr/ports/java/.
Click to expand...
Oh please. I am not 25 any more. Last time I checked MathWorks officially supported platforms were Red Hat, Ubuntu, Suse, Debian, Windows, and OS X. Oracle supports Solaris as well. Do you think I will jeopardize couple of million dollars worth of grants my Lab has to do real work by running our main tools on unsupported platform via binary compatibility layer :\ ? That is not how real world works. Until IBM, HP, Oracle, Google make the same kind of investment in FreeBSD they made in Linux and until there is Red Hat like entity take the ownership (I am familiar with IX Systems :) ) FreeBSD will remain to be what FreeBSD is. A great source of a good code for companies like Juniper networks and alike, a great solution for a small busnesses who use generic open source tools, fantasitc playground. No more no less. That is not necessary bad thing. BSDs and in particular FreeBSD would lose a lots of allure by becoming stiff, bloated proprietary OS which needs to run binaries from 25 years ago.
 
You only said that they don't RUN on FreeBSD. I just proved you wrong.
Oko said:
a great solution for a small busnesses who use generic open source tools
Click to expand...
You mean companies like Netflix, NetApp and until recently WhatsApp? ;)
 
Oko said:
1. There is a legal entity you can sue (Red Hat, Novel, Canonical come to mind)
Click to expand...
Try reading an EULA or two. You cannot sue anybody. Ever.
2. Vendor support. FreeMat, Scilab and GNU Ocatave are nice toys but unless MATLAB runs on FreeBSD my computing nodes will run Red Hat. The same goes with Oracle Java and many other common things.
Click to expand...
Support is the ONLY reason they're doing this.
 
pkubaj said:
He proposes that companies be responsible financially for money losses due to software they create. Every software creator would be responsible, unless they opensource their software.
Click to expand...
That's never going to happen. If I was a software company I would simply stop producing software due to the risk of liability.
 
SirDice said:
pkubaj said:
He proposes that companies be responsible financially for money losses due to software they create. Every software creator would be responsible, unless they opensource their software.
Click to expand...
That's never going to happen. If I was a software company I would simply stop producing software due to the risk of liability.
Click to expand...
Well, the solution is to opensource you software and only sell paid support for it. Take a look at what Red Hat takes money for. Canonical also wants to go the same way.
 
pkubaj said:
Well, the solution is to opensource you software and only sell paid support for it.
Click to expand...
You may not be allowed to do this. You may be using intellectual properties from some other company and not be allowed to open source it.
 
pkubaj said:
Every software creator would be responsible, unless they opensource their software.
Click to expand...
For a commercial product, I agree. I give you my product, you give me your money, and my product's value must match your money's value. So if my product is flawed, I am responsible.

But pretending warranties (and, consequently, responsibilities) for a non-commercial (even if closed-source) product that you have not paid and no one forced you to use, it seems to me a bit ungrateful and ill-mannered.

p.s.: are we still on-topic?
 
About the state of the Linuxator in FreeBSD, just take a look at the very small amount of commits done to the code during last few years. I can spot only a handful of commits that are related to implementing syscalls that were originally left unimplemented when the Linuxator was created.

The Linuxator is a cute toy that you can use to run the flash player on your FreeBSD box because there's no other option if you must have a working flash player. Outside that? Well, forget it.

http://svnweb.freebsd.org/base/head/sys/compat/linux/?sortby=rev&sortdir=down#dirlist
 
You must log in or register to reply here.
Back
Top