Virtualized storage server or bare metal? Pros/Cons?

[bexamous]

And as we all know, the Linux kernel is not the most stable thing out there. There are lot of sysadmins that consider Linux to be a toy OS, riddled with bugs and unstable.

... this is one of the most retarded posts I've seen in awhile.

 

[apnar]

Not to mention I wouldn't call ESXi linux. The vmkernel that runs as the hypervisor isn't linux based although they do apparently adapt some driver code from the linux base. You may be confused due to the Service Console that used to run as part of ESX which was Linux based.

 

[dustNbone]

This was my understanding of ESXi too. It runs Linux for the service console in a VM on top of vmkernel which afaik is not Linux at all. Is this different now?

 

[apnar]

ESX had a Service Console. They have now moved to only ESXi which gets rid of the Service Console VM and is pretty much only the hypervisor (although there is a very basic shell). They do offer a separate Linux based VM that you can run if you like that adds back some of the Service Console features.

 

[danswartz]

... this is one of the most retarded posts I've seen in awhile.

If I didn't have reason to believe otherwise, I'd think it was a troll.

 

[danswartz]

ESX had a Service Console. They have now moved to only ESXi which gets rid of the Service Console VM and is pretty much only the hypervisor (although there is a very basic shell). They do offer a separate Linux based VM that you can run if you like that adds back some of the Service Console features.

The VMA? Vmware Management Appliance (I think?) It's a fairly lean suse based app that provides the various vicfg/esxcli/etc commands via some perl libraries...

 

[brutalizer]

And a lot of sysadmins considered Linux to be far ahead of Windows for a very long time.

I am talking about Enterprise computers, Mainframe league, big Unix servers, etc. Not Windows servers, that is certainly a toy OS in this league.

As you know, Linux has a lot of distributions. Many of them are toy-ish distributions with little support for a "unique" and quirky home user.

I am not talking about distros. I am talking about the Linux kernel itself, which is considered as a toy, by big Unix/Mainframe admins. Compared to old mature Unix and Mainframes, Linux is unstable. There are many such sysadmins that would never let Linux into their server halls. Also, they would never let x86 servers into their server halls, as x86 cpu architecture is buggy and bloated. They only accept RISC cpus and Mainframes, and old mature Unix and Mainframe z/OS. Linux and Windows are not in the same league.

But of course, Linux is better and more stable than Windows. But both are considered toy oses. I mean, MS has built a xx billion company bigger than RedHat, does that mean that Windows is more stable than Linux? No.

Linus Torvalds calls Linux bloated and huge:
http://www.theregister.co.uk/2009/09/22/linus_torvalds_linux_bloated_huge/
"...Citing an internal Intel study that tracked kernel releases, Bottomley said Linux performance had dropped about two per centage points at every release, for a cumulative drop of about 12 per cent over the last ten releases. "Is this a problem?" he asked.
"We're getting bloated and huge. Yes, it's a problem," said Torvalds..."

There are many testimonies where people say Linux is buggy and unstable (in comparison to Unix/Mainframes).



Regarding virtualized or bare metal, the less points of failures, the better. Apparently virtualized is good for most use, except for the most demanding tasks. Thus, it seems to be more, matter of taste. Not matter of performance, nor stability. Both are good enough, for most common tasks. To me, the conclusion is:
-Both are equally good, w.r.t to stability - when we talk about common usage.

 

[iroc409]

Now this isn't making any sense. Who thought this was about mainframes, until I haphazardly mentioned it? Big iron is mostly a niche market, I'd wager most of enterprise uses PC-based servers. It's a relatively small portion of the market, and certainly not really part of the average virtualization discussion on Hardforum.

I've worked for places where AS/400 was the backbone of the company, but nobody sat around saying our Windows machines were toys.

EDIT: Well, if Wikipedia can be said to be reliable, my estimate might be a little off but apparently a good bit of big iron guys think Linux is A-OK (http://en.wikipedia.org/wiki/Usage_share_of_operating_systems#Mainframes):

Mainframes
IBM System z10.jpg

Nearly 95% of Fortune 1000 companies use IBM's Information Management System.[55]

Operating systems for IBM System z generation hardware include IBM's bundled proprietary z/OS,[55] Linux on System z and as at October 7, 2008; 3 years ago the prototype OpenSolaris for System z.

Gartner reported on December 23, 2008; 3 years ago that Linux on System z was used on approximately 28% of the "customer z base" and that they expected this to increase to over 50% in the following five years.[56]

Of Linux on System z, Red Hat and Novell compete to sell RHEL and SLES respectively.

Prior to 2006, Novell claimed a market share of 85% or more.
Red Hat has since claimed 18.4% in 2007 and 37% in 2008.[57]
Gartner reported at the end of 2008 that Novell had an 80% share of mainframe Linux.[56]

[edit] Supercomputers
Graph of supercomputer OS market share according to TOP500.[58]

The TOP500 project lists and ranks the 500 fastest supercomputers that benchmark results are submitted for. It then publishes the collected data twice a year. The November 2011 figures are below.
Source Date Linux IBM AIX Other Unix Microsoft HPCS 2008 Other References
TOP500 November 2011 91.4% 5.6% 0.4% 0.2% 2.4%. [59][60]

*Note: the table above is hard to read, but it says 91.4% of the TOP500 are Linux-based.

 

[brutalizer]

zLinux on Mainframes is not really stable. Some say it is a joke, and a marketing thing: IBM only wants to offer Linux on Mainframes, but it is not really usable. The IBM Mainframes are not using Linux for the critical things. Mainframes are used in the banking world, where you update lots of accounts with 10% batchwise. They mostly run COBOL, doing that. The zLinux stuff they are running today, are not the critical banking things. Partly it could be because a Mainframe cpu is much slower than a decent x86 cpu. This means that if you have a Linux workload suitable for cpu, then it will finish faster on a x86 than on a IBM Mainframe. In fact, the newest biggest fastest z196 IBM mainframe touting 24 of the the "worlds fastest cpu" at 5.26GHz, is outclassed by two 8-socket x86 servers. Therefore, the zLinux is not really taken seriously. The critical things, run COBOL on Mainframes. If you need performance for Linux related things, it is more efficient to run on x86 servers.



Regarding the TOP500 list, it says nothing. For instance, the IBM Blue Gene at ranking 7(?), is one of the fastest super computers so it should have superior hardware and software, right? Well, it uses 750 MHz PowerPC cpus. Thus, the cpu is not that really good.

These super computers are basically a cluster on a fast network, and Linux runs very well cluster based work loads: embarrassingly parallel work loads. To run a HPC cluster is easier than run a big SMP server. Such super computers run a stripped down, tailor made Linux kernel. It can only do one thing: calculate, and nothing else. It has not much common with the standard Linux kernel. It is heavily modified, and everything is thrown out. Highly specialized, built to do one thing. Thus, such a Linux kernel is not usable in a server doing Enterprise corporate work.

So, Linux running on clusters on TOP 500, say not much about the quality of Linux kernel. Not much common there.

 

[HammerSandwich]

...the newest biggest fastest z196 IBM mainframe touting 24 of the the "worlds fastest cpu" at 5.26GHz, is outclassed by two 8-socket x86 servers.

Is this true just for number crunching or for data throughput as well? I've been out of mainframes for a decade, but their data bandwidth when Syncsorting flat files always astonished me. Need to sort a few hundred thousand rows of a few hundred bytes each? Well under 1 second, IIRC.

 

[gjs278]

I am not talking about distros. I am talking about the Linux kernel itself, which is considered as a toy, by big Unix/Mainframe admins. Compared to old mature Unix and Mainframes, Linux is unstable. There are many such sysadmins that would never let Linux into their server halls. Also, they would never let x86 servers into their server halls, as x86 cpu architecture is buggy and bloated. They only accept RISC cpus and Mainframes, and old mature Unix and Mainframe z/OS. Linux and Windows are not in the same league.

you're blatantly making all of this up. emphasis on the bold mainly. that might be some faroff dream for a sysadmin in a galaxy far, far away, but anyone who says they would refuse an x86 server because "x86 is buggy" is just full of crap. if the company needs x86 to run a commercial product that only works on x86, or needs to release software for x86 systems, that same sysadmin who was refusing x86 servers will be racking them himself.

 

[hotzen]

you're blatantly making all of this up. emphasis on the bold mainly. that might be some faroff dream for a sysadmin in a galaxy far, far away, but anyone who says they would refuse an x86 server because "x86 is buggy" is just full of crap. if the company needs x86 to run a commercial product that only works on x86, or needs to release software for x86 systems, that same sysadmin who was refusing x86 servers will be racking them himself.

this. sysadmin's wet dream ...

Furthermore I don't understand this discussion. Mainframes á la IBM zSeries have the concept of LPARs for decades, to separate OS instances. I really don't think you're supposed to virtualize linux or windows instances on top of a full blown "enterprise" Mainframe.
This thread should discuss pros/cons of virtualization on the x86 architecture regarding win and *nix.
That mainframe-niche is a whole different story and has no connection to ESXi whatsoever.

Apart from that, a zero-downtime IBM zSeries is *the* sexiest system I have ever seen

 

[brutalizer]

Is this true just for number crunching or for data throughput as well? I've been out of mainframes for a decade, but their data bandwidth when Syncsorting flat files always astonished me. Need to sort a few hundred thousand rows of a few hundred bytes each? Well under 1 second, IIRC.

I have never talked about Mainframe data bandwidth. I talked about cpu performance. CPU wise, the Mainframes are quite slow. I/O wise, the Mainframes excel, because they have lot of I/O cpus, a big Mainframe can have 296.000 I/O channels. That is lot of I/O.

Here is an Linux expert, who ported Linux to Mainframes and compared Linux on x86 to Linux on Mainframes. He says that 1 Mainframe MIPS equals 4 MHz x86.

But todays x86 cpus, have 8-cores, each core running at 2.000MHz, that equals 8 x 2.000MHz = 16.000MHz. This translates to many MIPS. The biggest fully equipped IBM z196 Mainframe, gives in total 52.000 MIPS. It costs many 10s of millions, I bet. Maybe even 100s of millions?
http://www.mail-archive.com/[email protected]/msg18587.html

you're blatantly making all of this up. emphasis on the bold mainly. that might be some faroff dream for a sysadmin in a galaxy far, far away, but anyone who says they would refuse an x86 server because "x86 is buggy" is just full of crap. if the company needs x86 to run a commercial product that only works on x86, or needs to release software for x86 systems, that same sysadmin who was refusing x86 servers will be racking them himself.

Well, if you think I am making this up, about Linux considered buggy, and x86 considered buggy, I suggest you read my links. Here is one about x86, in short, it says that x86 sucks.
http://www.anandtech.com/show/3593
- The total number of x86 instructions is well above one thousand" (!!)
- "CPU dispatching ... makes the code bigger, and it is so costly in terms of development time and maintenance costs that it is almost never done in a way that adequately optimizes for all brands of CPUs."
- "the decoding of instructions can be a serious bottleneck, and it becomes worse the more complicated the instruction codes are"
- The costs of supporting obsolete instructions is not negligible. You need large execution units to support a large number of instructions. This means more silicon space, longer data paths, more power consumption, and slower execution.


Here are some more x86 bugs:
http://en.wikipedia.org/wiki/AMD_Phenom
You missed the excel division bug?
http://en.wikipedia.org/wiki/Pentium_FDIV_bug

this. sysadmin's wet dream ...

Furthermore I don't understand this discussion. Mainframes á la IBM zSeries have the concept of LPARs for decades, to separate OS instances. I really don't think you're supposed to virtualize linux or windows instances on top of a full blown "enterprise" Mainframe.
This thread should discuss pros/cons of virtualization on the x86 architecture regarding win and *nix.
That mainframe-niche is a whole different story and has no connection to ESXi whatsoever.

Apart from that, a zero-downtime IBM zSeries is *the* sexiest system I have ever seen

There are lot of stories of IBM Mainframes crashing. Sure, they dont crash often, maybe once in 5 years. But it happens.

There is an IBM Mainframe software emulator called TurboHercules.
http://www.tech-news.com/another/ap200601b.html

It is sold to Mainframe customers that can not afford another Mainframe in case of crash. TurboHercules runs on a x86 server, and fails over in case the ordinary Mainframe crashes. If Mainframes are that reliable, why is there a market for it? Why does IBM fight against a software emulator that hard?

If the Mainframes are that super fast as IBM claims, a software emulator should not threaten a real Mainframe, so IBM should not be worried? Emulating another cpu architecture is 5-10x slower. But still the x86 servers are real fast today so the emulation gives lot of performance, in par with a mid size Mainframe. So IBM is really worried about software emulators.

Tom Lehmann, co-founder of TurboHercules, wrote:
... We can run a reasonably sized load (800 MIPS with our standard package). If the machine in question is larger than that, we can scale to 1600 MIPS with our quad Nehalem based package, and we have been promised an 8 way Nehalem EX based machine early next year that should take us to the 3200 MIPS mark. Anything bigger than that is replicated by a collection of systems.

But now IBM has stopped TurboHercules. It is rumoured that the nick name "Big Blue", is from the discovery that IBM had more attorneys than the American government. They dont sit idle, doing nothing. No, IBM is using them very aggressively.
http://arstechnica.com/open-source/...patent-promise-targets-mainframe-emulator.ars

So, you can use the Hercules open source version, but not the proprietary TurboHercules version.