Please forget enterprise. It is used and misconstrued until it means nothing. Lets talk about manageability. How do you (easily) manage 1000 gentoo (or arch linux) servers? You could have a distcc farm to build your base distro from stage1 (if you needed to) or just copy down the binary stage 3 builds and then bootstrap using binary ebuilds, but it is still a whole lot more difficult than a full binary distribution such as Redhat or Debian. Dealing with large clusters of servers, the tools that they include or write and then open source are what really blow me back.
Just a few in no particular order:
- the RHEL kernel. Redhat has consistently topped the list of Linux kernel contributors for years. The first google hits for it were http://lwn.net/Articles/451243/ and http://lwn.net/Articles/507986/, but that hasn't changed for a looooong time. They basically have as much of a monopoly on core Linux kernel develops as is possible in such a large complex project. Quite literally, there isn't a company in the world with more Linux development chops than Redhat. If you run critical applications on Redhat servers (think banking or wall street exchanges like NASDAQ or hospital systems that downtime could result in real problems), Redhat will be able to fix it if anyone can. I'm not pretending working with Redhat support is fun, but they are better equipped from an engineering standpoint than virtually anyone. The numbers back that up. Due to this, the Redhat kernel is an interesting hybrid of slightly older and battle tested stable with newer features backported. This is achieved because super often the people who write the features upstream tend to be redhat employees, so they do both. If I was asked to pick one thing that set RHEL / Redhat / CentOS apart, it would be the work that goes into their kernel for QA and testing / backporting. Look at a company like Canonical, they have a bit more than a dozen (https://wiki.ubuntu.com/KernelTeam) kernel developers. They simply can't compete on engineering resources due to their limited number of engineers. As a result (and a smart business move) the are more consumers of patches from upstream than producers. Also, look at the lwn "who wrote linux X.YY" articles. You'll rarely and almost never find Canonical on that list except for when they got the apparmor patches merged (props to them!).
- sssd[1] - a solid implementation that essentially unifies pam ldap/kerberos, pam_ccreds/nslcd/nscd/pam_access all in one very nice implementation. This makes (for instance) joining your Linux nodes into an Active Directory domain (without using commercial software from likewise or some other cruddy vendor) just work out of the box. It also makes single-sign-on and migration from standard ldap to kerberized ldap (a very hard problem) super duper simple.
- cobbler[2] (and now the foreman[3]) - These tools along with redhat's kickstart make pxebooting a cluster of 500 new servers very trivial to turn into 500 new ready to use for production servers. Gentoo has nothing I'm aware of that allows installing completely automated like kickstart, but someone please enlighten me via a reply if this is incorrect. Michael Dehaan (big HN commenter and wrote the ansible config management tool) wrote cobbler.
- abrtd[4] / faf[5] - abrtd will collect crash reports (segfaults, coredumps, python tracebacks, kernel oopses, etc) and parse the info / store the relevant bits locally or forward on to a faf server. It will allow you to things (for example) like figuring out easily every single system that is reporting a specific kernel oops, which is then tracked down to a specific type of hardware and kernel combination. Sure there are tools like crash and netdump, but abrtd is simply a very modular management tool ontop of all of those things. The public fedora project faf is located at: https://retrace.fedoraproject.org/faf/problems/hot/. faf is good stuff
- freeipa[6] - Honestly up until this project, Linux never had anything that competed with Microsoft's Active Directory for a turn key easy to setup and manage kerberized ldap user and group / policy management product. IPA changes that and integrates very well with Microsoft AD through a kerberos level trust. sssd (above) is the ipa client. It allows true single sign on between Linux and Windows clients, something that is still elusive for most companies.
- standards. Linux's biggest strength is also it's achilles heel. Not having package standards or kernel standards (or stability) prevented a lot of companies from using Linux or certifying their software for Linux early on. Being very conservative in what they will support and supporting it for very long periods of time allowed companies like Oracle (as a horrible example) to port their database to Linux and certify that things are good. Try getting big complex commercial pieces of software working on a build your own distro. It is possible, but is buyer beware. Redhat made this their business model and has done a great job of it. At this point, Debian has also done a wonderful job at standardizing things and being consistent, albeit different, from Redhat.
TL;DNR: Redhat is building tools that make Linux easier to deploy and easier to manage in large "enterprise" environments. These tools make it equally easy to manage in smaller environments. No single entity has pushed Linux further in the "enterprise" than Redhat. I could list plenty more, but this hopefully answers your question fully. If not, click through to my profile, find my resume, and from it shoot me an email.
[1] https://fedorahosted.org/sssd/
[2] http://www.cobblerd.org
[3] http://theforeman.org
[4] https://github.com/abrt/abrt/wiki/ABRT-Project
[5] https://github.com/abrt/faf
[6] http://www.freeipa.org/page/Main_Page