Ask HN: Should I choose Sun Solaris or Linux?
I'm going to deploy into many servers in production i'm interested in performance & low cost support for multi-cores multi-sockets, high IO & CPU & Memory & Networking.
Should I choose Sun Solaris or Linux? if linux I thought of CentOS as its a free replica of RedHat.<p>Thanks
SunSolaris? Linux? FreeBSD?
10 comments
None of your requirements, at least when they're as unqualified as they are, can help you to pick an OS.
Ask yourself where your bottleneck will be, then ask yourself if the OS is going to make any difference there. The answer is very likely no.
No modern OS is going to give you any significant performance/cost advantage over another in serving regular web apps. Most of your trouble will be in setting it up and maintaining it, so go ahead and pick whatever you're most comfortable, or, if you're not comfortable, ask someone who are (and that you trust).
Also, don't pick CentOS, unless you need RH compatibility.
Ask yourself where your bottleneck will be, then ask yourself if the OS is going to make any difference there. The answer is very likely no.
No modern OS is going to give you any significant performance/cost advantage over another in serving regular web apps. Most of your trouble will be in setting it up and maintaining it, so go ahead and pick whatever you're most comfortable, or, if you're not comfortable, ask someone who are (and that you trust).
Also, don't pick CentOS, unless you need RH compatibility.
Why not CentOS? what would you recommend then?
I saw that oracle are really emphasizing how their OS has much higher performance than linux I wondered if its correct or if their benchmarks are good: http://www.oracle.com/us/solutions/performance-scalability/b... http://jp.sun.com/solutions/infra/eda/pdf/solaris-linux.pdf
So I wondered if i'm missing some tens of percents in performance by not choosing Sun Solaris...
I saw that oracle are really emphasizing how their OS has much higher performance than linux I wondered if its correct or if their benchmarks are good: http://www.oracle.com/us/solutions/performance-scalability/b... http://jp.sun.com/solutions/infra/eda/pdf/solaris-linux.pdf
So I wondered if i'm missing some tens of percents in performance by not choosing Sun Solaris...
CentOS is binary compatible with RedHat, so it's for people who need the enterprise-ness of RedHat, but can't/won't run RedHat. An example I've heard is running RedHat in production for the support, but CentOS for devs and stating. The reason to run RedHat is usually something like needing to run Oracle or a number of other very complex software suites (the keyword is complex, not better or faster) that are only supported on specific platforms.
If you're running Oracle on Sun hardware, go for Solaris, sure, if for no other reason the support. If you're running "regular" open source software on generic hardware, just use Linux/BSD. There's nothing in Solaris that'd make Postgres faster than Linux or BSD on the same hardware.
If you're running Oracle on Sun hardware, go for Solaris, sure, if for no other reason the support. If you're running "regular" open source software on generic hardware, just use Linux/BSD. There's nothing in Solaris that'd make Postgres faster than Linux or BSD on the same hardware.
What distribution do you recommend? (which is stable enough...) I wanted to choose CentOS because I know RedHat are saying they do lot of tests to their distribution before they release... so I wanted something that is fully tested and reliable.
It's been a while since I've had any responsibility over servers, but Debian has always treated me well.
Ubuntu which is derived from Debian is a good choice for servers, especially the LTS editions.
I'd be inclined to give it a try (I use and love Ubuntu on the desktop), but my impression is that Ubuntu is Debian packaged for the desktop... which isn't really relevant on a server. Debian is, OTOH, geared almost entirely for the server, already before Ubuntu came along.
I'd use FreeBSD.
OpenSolaris is officially dead. Solaris isn't low cost for sure. FreeBSD gives you a high performing io/network stack plus extras from Solaris world, like DTrace & ZFS, plus a focus of maintainers on just these features you seem to need most. This is contrary to many Linux distros, even commercial ones like RedHat, where the focus seems to drift to different directions over time, as they are trying to please different groups of users/donators/customers.
If you decide Linux anyway, I'd recommend Debian instead of CentOS. Contrary to common belief, RedHat/CentOS distros are "enterprise-oriented" maybe in the aspect of supporting typical Java development, but when it comes to production operations their enterpriseness does not necessarily shine, esp. when things break. YMMV.
These recommendations change a lot if you must deploy Oracle databases on the machines.
Well many people want to give their 2 cents in this thread; while this is excellent, I'd like to point I'm a sysadmin of various Unices since 15 years so I claim to know what I mean and not just promote personal prefs when saying something about OSes (if it was my personal bias I'd use Slackware forever). If you had further questions and wished to ask them privately, you'll find contact links on my profile.
OpenSolaris is officially dead. Solaris isn't low cost for sure. FreeBSD gives you a high performing io/network stack plus extras from Solaris world, like DTrace & ZFS, plus a focus of maintainers on just these features you seem to need most. This is contrary to many Linux distros, even commercial ones like RedHat, where the focus seems to drift to different directions over time, as they are trying to please different groups of users/donators/customers.
If you decide Linux anyway, I'd recommend Debian instead of CentOS. Contrary to common belief, RedHat/CentOS distros are "enterprise-oriented" maybe in the aspect of supporting typical Java development, but when it comes to production operations their enterpriseness does not necessarily shine, esp. when things break. YMMV.
These recommendations change a lot if you must deploy Oracle databases on the machines.
Well many people want to give their 2 cents in this thread; while this is excellent, I'd like to point I'm a sysadmin of various Unices since 15 years so I claim to know what I mean and not just promote personal prefs when saying something about OSes (if it was my personal bias I'd use Slackware forever). If you had further questions and wished to ask them privately, you'll find contact links on my profile.
If you are unsure of which to pick, perhaps it's indicative that you're going to have other problems ahead: maybe you should hire someone to advise you and help you out.
At the very least, the OP should talk to someone in person or on the phone about this. There are simply too many possibilities to adequately answer this for a production deployment.
Some questions that I (as a former SysAd) would ask:
1) How many is "many servers"? 10, 100, 1000, 10000?
2) What exactly is "performance for ... multi-cores multi-sockets, high IO & CPU & Memory & Networking"? This basically covers everything ever created. The high-performance for TeraGrid (sci. apps) is very different than E-Trade (financials) and very different from Amazon (e-commerce).
3) Is "low-cost support" $0 and 1 week of searching online forums? Or $100 and a phone call?
4) Do you like "port", "apt", "yum", "emerge", or "make"?
5) Do you need to have multiple versions of the same app available at the same time? E.g., gcc-2.97, gcc-3.2, 3.3.2, and 4.4.1?
6) Did you not know the answer to any of these questions? If so, I highly recommend you hire someone like @davidw says.
Some questions that I (as a former SysAd) would ask:
1) How many is "many servers"? 10, 100, 1000, 10000?
2) What exactly is "performance for ... multi-cores multi-sockets, high IO & CPU & Memory & Networking"? This basically covers everything ever created. The high-performance for TeraGrid (sci. apps) is very different than E-Trade (financials) and very different from Amazon (e-commerce).
3) Is "low-cost support" $0 and 1 week of searching online forums? Or $100 and a phone call?
4) Do you like "port", "apt", "yum", "emerge", or "make"?
5) Do you need to have multiple versions of the same app available at the same time? E.g., gcc-2.97, gcc-3.2, 3.3.2, and 4.4.1?
6) Did you not know the answer to any of these questions? If so, I highly recommend you hire someone like @davidw says.
My 2 cents: Oracle is going to drive Solaris into the ground (and OpenSolaris doubly so).
It's days were already numbered, like all proprietary Unixes, but while before I was betting on Solaris to be the last one standind, now I think it will probably be AIX.
Oracle is going to drive Solaris into the ground
On what do you base this? Larry is spending more on SPARC and Solaris than McNealy ever did! It's possible that Solaris will become less suited to (for example) HPC over time but for anything related to commercial data processing, including Java and web serving, expect for it to continue to occupy a leading position among the Unixes.
On what do you base this? Larry is spending more on SPARC and Solaris than McNealy ever did! It's possible that Solaris will become less suited to (for example) HPC over time but for anything related to commercial data processing, including Java and web serving, expect for it to continue to occupy a leading position among the Unixes.
"a leading position among the Unixes": that's the position that it already occupied: the biggest fish in a rapidly drying pond.
The killer combo of free Unixes on commodity (read x86) hardware cannibalized the proprietary Unix market. Where that trajectory will lead has been obvious for quite some time now.
Sun's measures to try to make Solaris survive that were a) port to x86 architecture and b) open source the OS. Jump out to a bigger pond, in other words.
Oracle's strategy with Solaris is going to be to squeeze every last penny from it, and when it dies it dies. Killing OpenSolaris and making it clear that open source releases of the code base will be done only "at some point" after the commercial release are just the beginning of that.
They will push Solaris in the same places where they currently push Oracle DB. Those places tend to be of very little interest to 99% of the HN croud. Java and web serving are not the kinds of things that justify stratospherically overpriced software licenses...
As for SPARC, I seriously doubt they will pour significant money into it. Trying to compete with x86 is akin to setting $100 bills on fire at this point. They will no doubt keep the Sun brand alive and use it to sell x86 based machines (which is the strategy Sun itself started of course).
Oracle's strategy with Solaris is going to be to squeeze every last penny from it, and when it dies it dies. Killing OpenSolaris and making it clear that open source releases of the code base will be done only "at some point" after the commercial release are just the beginning of that.
They will push Solaris in the same places where they currently push Oracle DB. Those places tend to be of very little interest to 99% of the HN croud. Java and web serving are not the kinds of things that justify stratospherically overpriced software licenses...
As for SPARC, I seriously doubt they will pour significant money into it. Trying to compete with x86 is akin to setting $100 bills on fire at this point. They will no doubt keep the Sun brand alive and use it to sell x86 based machines (which is the strategy Sun itself started of course).
"As for SPARC, I seriously doubt they will pour significant money into it. Trying to compete with x86 is akin to setting $100 bills on fire at this point. They will no doubt keep the Sun brand alive and use it to sell x86 based machines (which is the strategy Sun itself started of course)."
I believe they will. And they will also tailor it for Oracle DB and change the licensing plans to make SPARC a "wiser" choice, locking the customer both on SW and HW. The competition does it, so why shouldn't they do it also?
I believe they will. And they will also tailor it for Oracle DB and change the licensing plans to make SPARC a "wiser" choice, locking the customer both on SW and HW. The competition does it, so why shouldn't they do it also?
I still don't see a business case for it, though. Continuing to develop SPARC will cost a lot of money and won't buy them any extra lock-in. As far as DB apps are concerned Oracle is the platform, what it runs on is transparent. Edited to add: and the same goes for Java, which on Unix accounts for 99% of enterprise sw development.
As far as I can figure out the only reason IBM has to keep developing POWER is their mainframe business, which is the king of platform lock-in :)
As far as I can figure out the only reason IBM has to keep developing POWER is their mainframe business, which is the king of platform lock-in :)
POWER is bloody good, tho', check out the SPECInt results. If you have a workload you can't parallelize you can get double the performance of the best x64s.
"A leading position among the Unixes" is sort of like "a leading horse-buggy manufacturer" isn't it?
I don't see it "running into the ground" - people will be using it for many years in the future. However, the supply of people picking it for new projects probably just dried up quite a bit. People will 'grow up' knowing how to work with Linux, because that's what they tried at home, at school, what they learned on the job, and what their friends know. Those network effects mean that it'll mostly be sensible to keep switching to Linux over time, gradually, unless Solaris suddenly sprouts something that makes it a must-have, which I don't see.
I don't see it "running into the ground" - people will be using it for many years in the future. However, the supply of people picking it for new projects probably just dried up quite a bit. People will 'grow up' knowing how to work with Linux, because that's what they tried at home, at school, what they learned on the job, and what their friends know. Those network effects mean that it'll mostly be sensible to keep switching to Linux over time, gradually, unless Solaris suddenly sprouts something that makes it a must-have, which I don't see.
Well, no, not really. Just one example, Sun developed DTrace, a bit later Red Hat got SystemTap. Who's leading and who's following there? Sun has ZFS and Oracle has BTRFS and SGI released XFS, all of which are well in advance of ext3. So while it is fashionable to bash Unix for being old-fashioned and praise Linux for being "advanced", the reality is a lot more subtle than that.
I never said Linux was more advanced. What it is, is "good enough" and improving all the time: http://en.wikipedia.org/wiki/File:Disruptivetechnology.gif
Especially since it's easier to find people to work with it, more open, and not driven by a billionaire who wants you over a barrel.
Also, I think that Oracle is quickly putting paid to the sort of environment where innovations like dtrace happen, although I could be wrong.
Especially since it's easier to find people to work with it, more open, and not driven by a billionaire who wants you over a barrel.
Also, I think that Oracle is quickly putting paid to the sort of environment where innovations like dtrace happen, although I could be wrong.
You left out a big piece of information. Are you going to be setting up these servers physically (colocation?) or are you going to be going with a hosting provider?
Pick what you know, though from what you have posted I'm not sure you know either of the systems you mentioned very well. Actually, unless you are a professional systems administrator then even picking what you know may not be good enough. You can setup your own servers if you are running on a low budget, but ideally you would be paying someone to take care of all your administration needs.
So, I assume from your budget you won't be buying your own servers and even if you were interested in buying your own servers you don't have the expertise to set them up properly. You will then be looking at hosting.
You should probably be looking at a hosting provider which offers managed hosting. This will be your support. Don't worry so much about the distribution, just choose whatever that provider has available.
Arguably Solaris is superior to Linux in many ways. If you want to go that route I would suggest Joyent as a provider. Linux is wide open for hosting choices. Find one you like and is within your price range and go with them.
You probably need to start out with a VPS as opposed to a dedicated server. Done right, a dedicated server is generally much more expensive than a VPS. Also, a dedicated server requires a lot of knowledge just to know what to look for.
Your widest choice of offerings for a VPS is linux. Pick one which offers a managed plan and go with whatever that plan ships with. If you have a choice, just go with CentOS. Personally, I prefer Ubuntu, but that's not on your list.
Pick what you know, though from what you have posted I'm not sure you know either of the systems you mentioned very well. Actually, unless you are a professional systems administrator then even picking what you know may not be good enough. You can setup your own servers if you are running on a low budget, but ideally you would be paying someone to take care of all your administration needs.
So, I assume from your budget you won't be buying your own servers and even if you were interested in buying your own servers you don't have the expertise to set them up properly. You will then be looking at hosting.
You should probably be looking at a hosting provider which offers managed hosting. This will be your support. Don't worry so much about the distribution, just choose whatever that provider has available.
Arguably Solaris is superior to Linux in many ways. If you want to go that route I would suggest Joyent as a provider. Linux is wide open for hosting choices. Find one you like and is within your price range and go with them.
You probably need to start out with a VPS as opposed to a dedicated server. Done right, a dedicated server is generally much more expensive than a VPS. Also, a dedicated server requires a lot of knowledge just to know what to look for.
Your widest choice of offerings for a VPS is linux. Pick one which offers a managed plan and go with whatever that plan ships with. If you have a choice, just go with CentOS. Personally, I prefer Ubuntu, but that's not on your list.
I've used Solaris on Sun SPARC hardware and it is a very strong environment for heavy duty enterprise applications, especially if you have to use Oracle and/or Java. But very expensive solution. With Oracle now owning them, I don't think the cost of ownership is going to be coming down, but the quality of support might.
If you are using FibreChannel and SAN attached storage, then Sun might still be a good option. But seriously I'd look at HPUX and AIX as well.
My experiences with OpenSolaris on x86 (non-Sun) hardware has been disappointing. Maybe I had high expectations due to the SPARC experiences.
These days, I only use Ubuntu Server LTS editions. APT and the Debian infrastructure meets my needs well. Also use PostgreSQL rather than Oracle RDBMS.
If you are using FibreChannel and SAN attached storage, then Sun might still be a good option. But seriously I'd look at HPUX and AIX as well.
My experiences with OpenSolaris on x86 (non-Sun) hardware has been disappointing. Maybe I had high expectations due to the SPARC experiences.
These days, I only use Ubuntu Server LTS editions. APT and the Debian infrastructure meets my needs well. Also use PostgreSQL rather than Oracle RDBMS.
wearing dev,sysadm, architect hats running these platforms since 1999. i had the most consistency and general usability with freebsd. freebsd is easier to maintain and learn its features, docs are great. choose the one that you need to learn the least to run, install and maintain. yes i am somewhat biased to freebsd based on my own experience. the past say about 5 years solaris gave me many headaches like library dependecies, package management. freebsd, linux have good package managements. solaris has solid reliability and performance. all things considered, go with freebsd. freebsd security is easier to learn than selinux.
Interesting...
Wow, this is a question that can really fracture a community.
My two cents, from pretty extensive linux experience and one short Solaris experience: do what aligns more with your IT admin's philosophy.
My feel for Solaris is that everything is managed -- there are daemons to watch daemons whose job it is to make sure certain apps don't go down.
Linux is a bit less monitored -- "Oh, your apache process shouldn't have gone down? Shit you shoulda had something watching it."
Personally, I like the more ad hoc feel to linux. But pick whichever your sysadmin prefers. Solaris, to me, felt like I was being handcuffed.
My two cents, from pretty extensive linux experience and one short Solaris experience: do what aligns more with your IT admin's philosophy.
My feel for Solaris is that everything is managed -- there are daemons to watch daemons whose job it is to make sure certain apps don't go down.
Linux is a bit less monitored -- "Oh, your apache process shouldn't have gone down? Shit you shoulda had something watching it."
Personally, I like the more ad hoc feel to linux. But pick whichever your sysadmin prefers. Solaris, to me, felt like I was being handcuffed.
Why can't you deploy whatever you've been developing and testing on? That would make most sense to me.
For me, that would mean OpenSuse, but that seems to be a minority choice these days.
Anyway, I would advise against OpenSolaris after Oracle (paid for Solaris might be OK) - see the recent leaked emails etc. When I use CentOs I am always frustrated that they are using such old software versions (Python 2.4!), but that may just be the clients I have. For stability and security a BSD is not a bad idea, but again check for versions of the software you need.
For me, that would mean OpenSuse, but that seems to be a minority choice these days.
Anyway, I would advise against OpenSolaris after Oracle (paid for Solaris might be OK) - see the recent leaked emails etc. When I use CentOs I am always frustrated that they are using such old software versions (Python 2.4!), but that may just be the clients I have. For stability and security a BSD is not a bad idea, but again check for versions of the software you need.
Whatever you choose, stay clear of Solaris. Especially OpenSolaris (that's a dead-end).
If you're unfamiliar with both, Linux systems are easier to get used to, in my experience.
If you're unfamiliar with both, Linux systems are easier to get used to, in my experience.
So how come many "big" companies are using solaris with much success?
Many "big" companies are using Solaris (and other UNIXes) because they have been for a long time, and there is a cost to switching.
Upgrades to the kind of systems Solaris is being used (core systems and databases) tend to stay in the same course, not only because there is a lower risk in staying on the same OS/architecture combo, but also because most of those systems are old and tied to Solaris/SPARC. Unless there is a sufficiently strong reason to switch, most "big" companies prefer to pay the premium (in SW and HW) than assume the risk.
Solaris on commodity funcions is dying rapidly, if one can't already consider it dead.
The notion that Linux is less advanced is bogus. In some aspects Solaris is more advanced, in others Linux has the lead. What matters is what's the best choice for a each task given constraints of cost, reliability and availability of skilled technicians.
Upgrades to the kind of systems Solaris is being used (core systems and databases) tend to stay in the same course, not only because there is a lower risk in staying on the same OS/architecture combo, but also because most of those systems are old and tied to Solaris/SPARC. Unless there is a sufficiently strong reason to switch, most "big" companies prefer to pay the premium (in SW and HW) than assume the risk.
Solaris on commodity funcions is dying rapidly, if one can't already consider it dead.
The notion that Linux is less advanced is bogus. In some aspects Solaris is more advanced, in others Linux has the lead. What matters is what's the best choice for a each task given constraints of cost, reliability and availability of skilled technicians.
Legacy and inertia.
Historical reasons. Solaris was the definitive platform circa the late nineties. I used Linux back then, but it was definitely not ready for prime time yet. Many enterprises were built around Solaris. Today, Linux is a better alternative today for most use cases, but only marginally better. Until a year ago, it was not worth the cost of a switch.
Why a year ago? Oracle bought Sun. Sun was very good at creating open, useful platforms, but was very bad at making money off of them. Sun was bought by Oracle. Oracle isn't very good at building software, but is extremely good at selling it and maximizing profit from it. Oracle needs to make back the 7+ billion dollars it made on Sun somehow. How? We have part of the answer in the Google suit: by milking Sun's customer base.
Java has 10 million developers. To make 10 billion (there's a fair bit of risk in buying Sun -- if Oracle wants a 3x return, that's 10 billion each for Java, hardware, and Solaris), it needs to make $1000 per developer. How? I don't know how, but I'll run one possible scenario.
Step 1: Cut off competing JVM implementations with patent and copyright suits. This is already starting to happen.
Step 2: Release new versions of Java in two versions. The first is a free Java Express, which is okay for student use, but has no JVM, and omits many of the more enterprise-centric libraries. The second is a full-fledged version, and costs $1000 a pop. Developers would, of course, free to use the old, bit rotting version as well.
Most Java developers cannot switch platforms -- most are working on enterprise systems with hundreds of thousands or millions of dollars of sunk costs. Java immediately becomes a legacy platform, but even if only 50% of developers buy into it (even if only during the transition period), Oracle makes a cool $5 billion, almost paying for the rest of Sun.
Where does this leave Solaris? I don't know, but I suspect similar things are afoot. Oracle will want to make more money, and Oracle's customers will end up paying more. Oracle is already beginning to move in this direction with the death of OpenSolaris.
Aside from that, Oracle is incapable of attracting good developers. Former Sun employees are leaving as quickly as they can find new jobs. Will Solaris continue to have a technological advantage in the long term without Sun's brainpower? I seriously doubt it.
I know this is a lot of FUD, but with good reason. I am afraid, uncertain, and doubtful about the future of Java and Solaris. While I'm certain that Oracle will continue to develop Solaris so long as it continues to make them money, I don't know what they'll charge for it, or how well they'll develop it. I really wouldn't want to base my business around it.
(A secondary, technological answer is that Solaris advantages are much better suited to the climate of the nineties than of the teens. In the nineties, we tend to rely more on big computers with many CPUs, massive amount of memory, and many hard drives. In the teens, the trend is much more towards big arrays of cheap, unreliable, redundant boxes. Solaris' advantages are much more pronounced for the former than the latter.)
Why a year ago? Oracle bought Sun. Sun was very good at creating open, useful platforms, but was very bad at making money off of them. Sun was bought by Oracle. Oracle isn't very good at building software, but is extremely good at selling it and maximizing profit from it. Oracle needs to make back the 7+ billion dollars it made on Sun somehow. How? We have part of the answer in the Google suit: by milking Sun's customer base.
Java has 10 million developers. To make 10 billion (there's a fair bit of risk in buying Sun -- if Oracle wants a 3x return, that's 10 billion each for Java, hardware, and Solaris), it needs to make $1000 per developer. How? I don't know how, but I'll run one possible scenario.
Step 1: Cut off competing JVM implementations with patent and copyright suits. This is already starting to happen.
Step 2: Release new versions of Java in two versions. The first is a free Java Express, which is okay for student use, but has no JVM, and omits many of the more enterprise-centric libraries. The second is a full-fledged version, and costs $1000 a pop. Developers would, of course, free to use the old, bit rotting version as well.
Most Java developers cannot switch platforms -- most are working on enterprise systems with hundreds of thousands or millions of dollars of sunk costs. Java immediately becomes a legacy platform, but even if only 50% of developers buy into it (even if only during the transition period), Oracle makes a cool $5 billion, almost paying for the rest of Sun.
Where does this leave Solaris? I don't know, but I suspect similar things are afoot. Oracle will want to make more money, and Oracle's customers will end up paying more. Oracle is already beginning to move in this direction with the death of OpenSolaris.
Aside from that, Oracle is incapable of attracting good developers. Former Sun employees are leaving as quickly as they can find new jobs. Will Solaris continue to have a technological advantage in the long term without Sun's brainpower? I seriously doubt it.
I know this is a lot of FUD, but with good reason. I am afraid, uncertain, and doubtful about the future of Java and Solaris. While I'm certain that Oracle will continue to develop Solaris so long as it continues to make them money, I don't know what they'll charge for it, or how well they'll develop it. I really wouldn't want to base my business around it.
(A secondary, technological answer is that Solaris advantages are much better suited to the climate of the nineties than of the teens. In the nineties, we tend to rely more on big computers with many CPUs, massive amount of memory, and many hard drives. In the teens, the trend is much more towards big arrays of cheap, unreliable, redundant boxes. Solaris' advantages are much more pronounced for the former than the latter.)