Solaris vs. Linux
Copyright 2005-2012, Dr. Nikolai Bezroukov. This is a copyrighted unpublished work. All rights reserved.
- Most of the paper was written in 2005-2006, long before Oracle acquisition of Sun and subsequent closing of Solaris code. With Oracle acquisition the advantages and disadvantages of Solaris changed. Moreover, now those two OSes do not compete in the same league: one is closed source the other is open source. Although «self-closing» of linux distributions such as RHEL and SUSE due to complexity is evident to any observer of the field, htere is advantages of access to source code such as security inspections, etc. References to consequences of Oracle acquisition and elimination of Open Solaris were added only as afterthought and they do change the equation in favor of linux, despite its warts (which are evident for any specialist) and all the architectural shortcoming. As Oracle now concentrates on midsize and high end Intel CPU based servers, part of the paper devoted to workstations and low end servers became by-and-large obsolete.
- Readers with strong allergy to grammar and syntax errors should probably avoid reading this paper until later, supposedly more polished, versions. The current version is pretty raw. English is not the native language for the author. Treat it is as a preprint
- This is not a scientific comparison per se, but more of an attempt to formulate (and for the author to understand himself the issues involved 😉 the framework for comparison of two complex and successful OSes. This framework might be used by future researchers of this problem as well as business decision makers.
- Please pay attention to the version posted. This paper covers results of my own research and is work in progress. If you consider a particular part of the paper biased/incorrect please provide feedback. Opinions of the author change as he deeper understands issues involved in the process of this complex research. The topic is too complex and some errors are inevitable in a paper written by a single author. Actually the fact that a particular person is interested in the theme to the extent that he/she wrote such a paper is a strong argument against his/her depth of knowledge in one or both OSes that he/she tries to compare ;-).
- As large enterprises are slow adopters of new technologies and the paper is a volunteer effort it can be substantially behind the curve and most results are applicable only to enterprise versions of OSes that were current in 2006-2007. This was the time when the bulk of the paper was written. Amount of time that the author can devote to research of the topic was very limited. Some recent developments such as the enterprise success of Oracle Linux as a clone of RHEL with some own ideas (version of kernel, BTRFS filesystem, etc) is not reflected in the paper; Solaris 10 and Solaris 11 developments after the acquisition by Oracle are also not fully reflected in the current version of the paper. For example discussion of increased value of ZFS due to stabilization of the code and emergence of solid state disks as enterprise class hardware is also missing.
When you open your Windows
you’ll see a light blue sky filled with clouds.
If you look past the clouds, you’ll only see the Sun.
— Alan Orndorff
«The rumors of SPARC’s death have been greatly exaggerated»
USENET sig
Contents
Abstract
The key research area of the paper in not only comparative advantages of Solaris vs. linux on Intel platform, but also the cost of Unix flavors proliferation. The paper carries a strong «counter-proliferation» of Unix flavors stance. Sometimes I probably am too harsh about linux, but that’s just because in real life it makes sysadmin environment in large enterprises more complex then it was before its introduction just by adding two additional flavors of Unix to the current enterprise mix (start-ups and small firms are in completely different position here and most of the critique is not applicable). when Linux is secondary OS and Solaris is primary there is a definite bias toward Solaris and vise versa. That’s life.
Excessive variety of Unix flavors in large enterprises is a widespread disease that is a sign both technically and politically weak IT management. And absurd level of incompetence among them is not uncommon: I’ve personally known one manager of Unix group of a large corporation who did not understand that Intel CPU and Power CPUs use different instruction sets and as such need different binaries ;-). Such despicable situations are also a sign of IT losing political influence within the large corporation hierarchy of departments. Which lead to delegation of bean counters into management positions.
I think that interests of system administrators should be given a proper consideration in case of any new OS introduction in large enterprise environment. And if introduction of new Unix favors does bring savings part of them should be used for retraining. For me that means that if we need to add one flavor of Unix, we also need to remove one (or better two 😉 existing flavors: the best large corporate configuration is just two flavors of Unix on the floor as it is dangerous to put all eggs into one basket. With the current complexity supporting more then two flavors of Unix is suicidal. This means adding linux in reality should mean killing HP-UX, AIX, or even Solaris. As the hero of famous O. Henry story «The Roads We Take» quipped in not so different from large enterprise IT circumstances «Bolivar cannot carry double.» 😉 [O.Henry, 1910]. The article warns potential adopters of linux that it is difficult to capture value by just adding a new rider on the system administrator horse independently of which operating system we are talking about. And in case of linux we need to talk about two major enterprise personalities: RHEL and SLES, which are replaying Unix wars on a new level (with Ubuntu in the shadows). So when one adds new flavor of Unix one existing flavor needs to be killed. In case of linux probably two should be killed to maintain sanity. That leaves IMHO only linux and Solaris on the floor 🙂
The paper also introduces a generic framework of OS comparison that can be used for comparing other Unix flavors with Linux as well as between each other. It rejects simplistic approach dominant in comparative reviews of enterprise Unixes. In all such discussions it is paramount to separate tremendous price/performance advantages of Intel hardware from the advantages of the OS. As currently the only other enterprise flavor of Unix available on Intel hardware is Solaris X86 the paper is limited to comparison of those two Unix flavors but the methodology itself is more generic.
It is important to understand that operating systems kernels are a side show of open source movement or open distributed collaboration movement in a broader sense. Scripting languages and applications (especially scripting languages) are the key components of the open source movement and at the same time enablers of other forms of open collaboration exemplified by Wikipedia. So it’s Perl, PHP, Python, Apache, Jboss, bind, postfix, Postgress and MySQL that are flagships of the movement with LAMP stack (for Linux, Apache, MySQL, and PHP/Perl/Python) as open source highest achievement. I would like to stress it again that the essence of the movement is in utilizing and, simultaneously, enabling distributed collaboration. Many Linux enthusiasts make a typical for technological utopians error. They think that their beloved OS due to this wonderful thing called distributed Internet development will wipe away all traces of its predecessors � as if when you change the fuel you change the car. That’s proved to be false: Linux is and will remain just a flavor of Unix with its own (multiple) strengths and no less numerous weaknesses.
In a way Linux kernel is a new «SuperBIOS» on which open applications run and from the technical standpoint it has all the attractiveness of the good old BIOS (which is another way to say that it’s pretty boring). Moreover, Linux is just one of many interesting open source kernels, which (for unrelated to its technical merits reasons) is the most popular and gets all PR ink. FreeBSD, OpenBSD and other BSD kernels are similar in technical capabilities, less bloated and have some subsystems (FreeBSD) or packaging (OpenBSD) that are more technically interesting. In many cases they can substitute Linux in LAMP stack and few will ever notice the difference. Actually the difference can be positive: FreeBSD pioneered jails — a light weight virtual machine in 1999; OpenBSD pioneered integration of SSH and firewalls and defines the state of the art of secure X86-based Unix; NetBSD still defines the standard in Unix portability although Linux recently doing pretty well in this category too.
When one tries to provide a sober assessment of relative merits of those two OSes in large enterprise IT environment and concentrates on the nonpolitical, real world issues of administering multiple flavors of Unix servers one fact emerges as really startling: the current complexity (or, should we say, overcomplexity) of Unix environments lead to the situation in which most sysadmins can master well just one flavor of Unix. Better one might be able to support two (with highly asymmetrical level of skills, being usually considerably more proficient in one flavor over the other). And that’s it. Sysadmin who supports three flavors of Unix is by definition a mediocre sysadmin as he has little chances to develop real mastery of any of those three flavors. That leads to natural specialization inside Unix department when each specialist is responsible only for its «own» flavor. Needless to say that this is not very healthy situation that often results in internal fractures, infighting and other corporate ills.
In those circumstances Linux often acts as a spoiler and tends to increase, not decrease the variety of Unixes in the large enterprise environment. Moreover Linux itself is suffering from famous Unix curse with Suse and Red Hat competing in enterprise environment (while rare in the USA Suse is more popular in Europe and large European multinationals). That’s why positive effects of linux introduction without removing one of existing flavors of enterprise Unix (which in large enterprise environment is very difficult to do due to acquisitions) are much more questionable for large enterprises then for startups. That does not mean that adoption is slow or benefits are not existent. It just mean that financially linux introduction does not bring substantial savings, if Solaris 10 on Intel is already used in the environment. It bring other advantages such as ability to use open source ecosystem, and savings due to cheaper hardware, but as an OS the cost of support is similar. In few cases when linux is used on IBM Power architecture it can actually increase the costs. While linux is over 20 years old, not everything can be run on linux both for both technical (stability) and political reasons. That means that despite having important «home field» advantage for open source software deployment, the saturation point for Linux in large corporate infrastructures exists and might be lower then generally assumed. Few IT managers want to put all open source eggs into one basket.
The author argues that despite linux popularity and the role of standard de-facto for open source applications development instead of addition of linux one can add Solaris on Intel as it runs on the same high price/performance ration hardware and typically is already present in enterprise environment. Or you need to cut one existing flavors of Unix to preserve the current balance of enterprise OSes.
The paper was written with the explicit goal to serve as an antidote to primitive reviews on Linux self-congratulation sites styled like «I found old PC in the closet, dusted it off, tried to install Solaris on it; my God what a crap Solaris is in comparison with Linux». Such reviews are not only misleading, they disorient open source enthusiasts (especially among staff of large companies) conditioning them against a more stable and in several areas (virtualization is one; Java applications performance is another) more advanced server OS that has a lot to offer. The role of paper as an «antidote» to overselling of linux in enterprise environment somewhat influenced the style making the paper more polemic, then it probably should be. Please note the author consider Linux and Solaris to be two best enterprise OSes despite the fact that paper does not specifically criticize AIX of HP-UX 😉
One of the non-trivial conclusions of my research that led to the writing of the paper is that the total cost of ownership of Unix as an OS class is highly correlated with the number of Unix flavors used: increasing the number of Unix flavors used in large enterprise datacenter typically results in increasing total cost of ownership independently of which flavor of Unix you are adding: Solaris, Linux or something else.