TBray Response: Sun Should Stop Sucking | Hacker News Reader

TBray Response: Sun Should Stop Sucking

kirkwylie.blogspot.com

39 pointssoundsop37 comments

Sun want to sell servers. They can either become a commodity box-shifter on razor thin margins, or they can make their products better so customers want to pay a premium.

To make their boxes better, they'd have to come with pre-installed server software, all the normal stuff you'd see on a web server, and tightly integrate the OS / file system / database so that (e.g.) MySQL is running very fast.

Kirk Wylie is right that no-one cares anbout Solaris anymore. If they'd open sourced it ten years ago, things might be different. But today, Linux is king on servers. I use Ubuntu on the desktop; I use it on the server too out of convenience -- I know that i can install new package on both systems in exactly the same way using apt-get, I know that the diorectory structure is the same, I know that I can test a feature on my desktop then port it to the server without getting wierd incompatibility problems.

The only way I'd consider using Solaris is if it worked exactly like Ubuntu, i.e. make the Solaris kernel a plug-in replacement for the Linux kernel in the Ubuntu distro. Fortunately, such a thing exists: http://foss-boss.blogspot.com/2008/11/nexenta-can-you-say-so...

Then, in order to be a one-stop shop, Sun should get into the hosting business, setting up datacenters around the world running their hardware/software solution. They could even do a system that automatically mirrored a web server at different locations around the world to make it more resilient.

On servers, Solaris is better than Linux. DTrace, ZFS, Zones, resource management framework, RBAC, service management framework, among other things, are all better. Solaris already has most of the Linux userland available. A lot of it is even installed by default; it just isn't in the default PATH. If you change your default path, you can get access to a lot of the inferior Linux tools that you are used to.

Yum and apt-get are better at patching and package installation than Solaris tools. However, the experience on Solaris isn't so bad that the whole OS has to be thrown away because if it. Besides, when IPS is released Solaris's package management will be better than both Yum and apt-get.

> "Solaris is better than Linux."

That's a very blunt statement of fact.

Yes, DTrace is great. ZFS has a lot of greatness if you overlook the bugs production systems have had (see: Joyent) to fix still (systems down for a week while they hand-fixed the FS - Yeah, fun stuff). Zones are cool, but has its drawbacks (and advantages) when compared to something like Xen.

> "A lot of it is even installed by default; it just isn't in the default PATH."

Uh, you know that when people talk about "Linux" they really mean "common GNU/Linux distributions such as Debian/Ubuntu/Fedora/RHEL", right? You realize they mean more than gmake and gfind and gcc, right? Sorry, Solaris has absolutely nothing like the apt repositories available on Ubuntu (I know they're working on this, but that is still the not-so-stable-or-bug-free OpenSolaris, and it still isn't anywhere close, and we're talking about today, right?). Can I pull down svn, git, mercurial, bzr in a simple command? How about lighttpd, nginx, haproxy? What about a recent vim and emacs? Recent memcached, python/perl/ruby client for it? Those are all basic and well known, that isn't even the hard stuff. What about something a little more specific? PostGIS, GDAL, Proj? I can grab all of those on Ubuntu in a few seconds.

And of those you can compile from source yourself on Solaris, you'll find many a quirk because the project maintainers tend to run either a flavor of Linux or FreeBSD.

Yeah, Solaris is "better than Linux on servers", and my time isn't worth anything.

...and Solaris doesn't even ship with GNU grep or find! I have to do crazy perl gymnastics to replicate grep -C (though that's mostly because our crazy systems group refuses to allow non-Sun binaries).

I can grab all of those on Ubuntu in a few seconds.

You're absolutely right, but when it's time to do something serious - in the Cal Henderson, irony laden, finger quote sense - you'll recompile them just the same.

I will?

Sorry, I've never bothered to recompile svn or python or many other things that my Linux install comes with. And I've use them every single day. For money. To feed my face hole.

Yes, when the time comes I can recompile anything on Ubuntu, but you know what? I'll take their already existing package, add patches, and easily create a .deb I can distribute among all the machines I use. Based on the existing package, with existing Ubuntu patches added that fix existing problems for me. Rather than going from scratch all alone in Solaris land.

For the purposes of this discussion nothing you've mentioned falls under the serious banner. I mean no snark at all in saying that - just that those with half a dozen 12 CPU boxes packed with nearly 200Gb of RAM tend to be cutting their own path.

On servers, Solaris is better than Linux

In terms of the quality of the software this may well be the case. However I've got lots of experience with Linux (especially Ubuntu) and little with Solaris. Familiarity means I can develop more quickly on Linux than on Solaris. T|ime is money. Even if, on the same hardware, Solaris runs faster, with the money I've saved on quicker development times, I can buy more expensive (and therefore faster) hardware.

BTW, of the programs you meantion, the only one I've ever heard of is ZFS; I've no idea what they others are or what they do -- maybe Sun need to be better about getting the message out on how good their system is.

By putting "Zones" in that list, you've given me the impression that you aren't really thinking this through, but rather are just parroting the Sun feature-function-benefit table.

Are you really arguing that Sun has Linux beat on virtualization and isolation?

For some classes of problems, Sun has a better, more robust solution in the form of Zones.

Sun also has Xen and VirtualBox (not as good as VMWare yet).

If you don't like Zones, fine. However many people are using them in a way that serves their needs - but perhaps not yours.

Zones are far less robust than Xen or VMWare. They're a short step up from chroot. Any Solaris kernel problem takes down all the zones, and there are a number of ways zones can influence other zones.

The flipside of that is update-on-attach, which Xen doesn't have.

My answer is the same as the last time you asked me:

http://news.ycombinator.com/item?id=318592

That was a different discussion, where I was simply pointing out the reasons I didn't like Zones. But this discussion is harder for Solaris to win, because you have to line Zones up against the very real and powerful virtualization systems Linux already has, like Xen.

We consult all the time for F500 companies with pools of Solaris boxes. Some of them use Zones. None of them virtualize Solaris. All of them virtualize Linux, and can move entire Linux servers from machine to machine from a UI. It is hard to convince me that Solaris has any kind of edge in the server market.

I sincerely would love to hear of the concrete problems with Solaris Zones, because I am deploying a system based on Solaris Zones very soon.

I don't think security is an issue--it is true that there could be a security bug that would let some malicious code escape from a non-global zone to the global zone. But, AFAICT, that could happen with Xen too. Plus, if there is malicious code executing in a non-global zone then I'm already screwed since that is where my applications and data are.

Xen's live migration requires shared storage (AFAICT), which I don't have. That means that for my systems, Xen migration would work exactly the same as Zones migration (shut down the guest, move it to the new system, and start it up).

I chose Solaris Zones because (1) Zones have extremely low resource consumption, (2) they are extremely simple to manage, (3) they are extremely simple to backup and restore when the zone is on ZFS, and (4) I get to use all the Solaris manageability features.

Let's start with I have a documented security finding in Zones; I'm not sure if it's been published and patched yet, and I'm not pointing it out, but you can track me down on LinkedIn or, I don't know, a lot of other places if you need bona fides.

Then think about the difference between Xen and Zones:

Zones share a kernel.

Xen machines do not.

This has two security implications:

* It is possible for there to be bugs that will allow one zone to see kernel resources (files, sockets, pids, IPC descriptors, etc) from another zone. It is impossible for that to happen with Xen, because VMs don't even know about each other.

* The majority of Solaris (and Linux) kernel vulnerabilities in the last several years have been kernel vulnerabilities, which Zones don't protect you from. The Xen hypervisor is significantly smaller than the Solaris kernel, and has had a far better security track record.

Server != web server. I think that's something that a lot of "Linux has won" people miss.

In the scientific computing world, the move away from Solaris was swift and comprehensive. It's not a question of what's better, but of what gives you the best bang for the buck at an acceptable level and Linux has more than delivered that.

There seems to be some Sun-focused FUD and Linux-focused fanboyism flying around here and on Kirk's blog. Obviously, that's no way to evaluate competing solutions. We need facts. We need apple to apple comparisons of hardware and operating systems, where we can get them.

Of course, every /serious/ business can and should get into the details, but for the sake of argument, I invite you to compare a couple apples from the barrel: Sun Microsystem's Sun Fire X4150 (used by Kirk's site), and Dell's PowerEdge 1950 III.

They support /very/ similar hardware options (rack units, CPUs, memory, network, internal storage, out-of-band management, power, expansion slots, etc.), have roughly the same price depending on options (~$7500), and run roughly the same major software (Linux, Solaris, VMWare). Based on that one example, I don't see how Kirk justifies decrying Sun.

I could go on with more hardware or a light comparison of Solaris vs. Linux, but you get the point. Stop with the hyperbole and try presenting facts in a digestible format.

After pointing out that the author of the article is too biased in favor of Linux to do an honest comparison, you then base your substantive critique on the argument that Sun hardware really isn't more expensive than the X86 hardware web shops use to do the same job.

My question is, do you actually believe that?

I wonder if the high prices are something we "normal" guys simply can not understand. Maybe they just trigger the "reassuringly expensive" button at companies that don't care about money. Or there are some complicated tax evasion schemes or whatnot at work, that make it mandatory to waste some money.

I've witnessed it for years in the Java world where every application had to run on Bea Weblogic and Oracle, when Open Source applications would have been available and actually were more suitable to the task.

Also, it seems to work for Apple...

From a marketing standpoint, setting your product prices much higher than the competition is one way to signal customers that your product is superior. But this tends to be more effective with consumer luxury products where buying decisions are based most highly on emotion and desire for exclusivity.

"Also, it seems to work for Apple..."

I see irony here, in that I'm old enough to remember when Sun was killing because everyone needed high end server hardware to get on the web and so Sun could charge a good premium, and Apple was getting killed because they were trying to sell a premium product into a market that had been commoditized by the PC.

Interesting how things change. I'm not sure what the equivalent of the "Apple strategy" would be for re-establishing a premium brand in the server space.

Correction: everyone thought they needed

I was around during the dot-com bust, and this little piece of revisionist lore always irks me. Something called "Linux" existed in 1998-1999, but it is not the "Linux" you know today, and when stacked up against Solaris running on Sun hardware, it didn't really work. People wanted to run their entire back-end on a single system, and that single system pretty much had to be a Sun E10000 running Oracle, because nothing else could scale up as high. The LAMP-stack scale-out style of system-building had not been invented yet.

I agree with you that people liked scaling using a single big machine back then, and Sun had the biggest and best machines (and still does to some extent).

The problem is more that a ton of those people didn't actually need to scale.

No-one planned not to need their servers...

> Maybe they just trigger the "reassuringly expensive" button at companies that don't care about money.

Maybe in the past, but do you think they're going to find a lot of companies that don't care about money in the near future?

I think there will always be companies with too much money. It wouldn't be my preferred strategy, but still...

Among other things, don't they say it is easier to raise a lot of money than just a little bit of money from VCs? So if you business plan lists 10 Sun servers instead of 10 cheap Linux servers, you already increased the money at stake by a large margin (I presume).

Sun is just a bunch of really smart developers surrounded by a bunch of fairly bozo sales and marketing folks, and headed by no one with a real vision.

They might as well give up.

"Without a vision, the people perish." (quoting scripture)

Sad, too, as I remember the early Sun quite well (at the 10-15 people phase), as they were a neighbor of ours in Mountain View, and we (Imagen, later QMS) swapped early laser printers for early Sun workstations.

Misquoting a mistranslation of scripture, unfortunately. (The actual proverb seems to mean something like "without prophecy, the people cast off restraint", which is both less poetic and less sensible.)

I think Sun should offer Sparc chips at half the price of Intel chips.

everyone keeps talking about what sun should or should not do. no one talks about how they will actually do any of it. sun is barely surviving; they don't have the money to do much.

Sensible response.

Is it?

The rebuttals don't seem to align at all with Tim's statements... And the "What Sun Could Do" seems to boil down to: sell more hardware and storage... Doesn't seems like a killer strategy to me.

Maybe they don't have a "killer strategy". They've been struggling for quite a while. It seems pretty plausible that the best they can hope for, for now, is a "survival strategy".

Sell more hardware, stop wasting money, and keep your customers happy.

Reply on news.ycombinator.com