Re: [aclug-L] ntmag

["Greg House" <ghouse@xxxxxxxxxxxxx>]

-----Original Message-----
From: phrostie <phrostie@xxxxxxxxxxxxx>
To: aclug-l@xxxxxxxxxxxx <aclug-l@xxxxxxxxxxxx>
Date: Monday, May 03, 1999 4:02 PM
Subject: [aclug-L] ntmag

>a man i used to work with in austin is considering learning linux.
>his day job is as an nt administrator.
>today he sent me this article asking me what i thought .
>in truth most of what the author is talking about is over my head.

I understand what the guy's writing about, and from what I know of Linux internals (not a lot), I'd guess most of what he's saying is legitimate enough.  What he ignores is the simple fact that while Linux may not scale to the high end enterprise level hardware that Solaris will, it performs outstandingly well in the nitch it actually fits in. 

As far as I'm concerned, this article is FUD simply because it wants compare dissimilar things and has a whole lot of technical mumbo jumbo to confuse those who don't understand such things into believing that these issues are revelant. Here's my point; Linux has never been put forward as an ENTERPRISE LEVEL solution!!!  It's primary positioning in the OS world is for a single or dual processor Intel-based server, where is performs admirably. My Volkswagon van carries a lot of junk, a lot more then a Ferrari. But it doesn't go very fast, and doesn't pretend to. It's useful for carrying things around. A Ferrari is great for racing, but it'll getcha a ticket faster then anything if you try to drive it around town.  And you can't even put a guitar amp in it, much less an entire band's worth of musical equipment.  The point is that Linux wasn't designed to run on a 64 processor system.  None of the developers HAVE ONE to work on.  I'd guess that if some kind soul wanted to give someone like David Miller or Alan Cox a 16 processor system to do development on, you'd start seeing some really nice support for that kind of thing.  but the fact is that these people are all volunteers (some paid volunteers, but not many get their hands on big iron...).

It's also FUD in that very few people really NEED an 8 processor system (which the 2.2 kernel supports), and the fact that while NT may support it, it's not NT's forte either.  This article wants to compare these operating systems by internal technical details rather then real world performance results.  Other then that very questionable Mindcraft study, every benchmark I've read for server performance (SMB/HTTP) has place Linux (Samba/Apache) at 2-3 times the throughput of NT on the same hardware.  These are facts, not theory.  It's based in the real world, not in someone's critique of the OS's internals.

You asked to avoid NT bashing, but there's a reason it has the reputation it has for crashing a lot...it crashes a lot (or hangs a lot, or has applications that hang a lot and you can't reset them without rebooting...).  I find it particularly ironic that the author of the article wants to talk about scalability issues, since NT seems to have the worst problems with that.  It seems to work fine when you're not pushing it's capability, but when you start really expecting some performance out of it, it starts hanging, crashing, etc.

The Blue Screen of Death is a common sight for an NT administrator with a busy machine.  As are the messages warning users that you're going to have to reboot to get the Exchange server responding again.  It appears that they have to do our mail server at work about once a week.  I wouldn't bet my enterprise on that.  It doesn't have any built-in remote administration tools either.  You think I'm gonna set something up on a massive multiprocessor machine costing multiple hundreds of thousands of dollars that I can't dial into to diagnose a problem?  The answer for something enterprise ready is absolutely NOT "Well, something's not working, we'll have to reboot."  The kind of machines in question typically cost their owners thousands of dollars in lost business every minute they're down. It'd be outstandingly foolish to pick NT to run that.  It's not very fast, it's not very stable, you can't get good support out of Microsoft.

Frankly though, I wouldn't bet my billion dollar system on Linux either.  If I really had something that required that amount of processing power, I'd buy something where someone would jump through hoops for me if it had a problem.  Believe me, if someone spends that much money on that kind of hardware, the vendor will go out of their way to make sure the support is exceptional.  It's easy to say a vendor like Sun doesn't give great support for a single low-end workstation, but buy an Enterprise 5000 and I think your response will be different (especially if they think you might buy another one...)

>my understanding is that linux has always had support for SMP,
>with 2.2 it is more efficient.

As someone else said, the 2.0 kernel had simple SMP support, 2.2 improves on it.  Apparently it still has a way to go.  I don't see that as a problem.  Consider Linux's lineage and development model and it's not hard to understand.  Again, he wants to compare apples to potatos.  Linux never claimed to be an enterprise class OS, just a lightweight kernel with some nice tools available for it.  The fact that it blows NT off the roadmap for a lot of applications is kind of amazing in itself.  It never tried to compete, it just did.

>i would like to give him some constructive feed back, but
>once again this is over my head.

Constructive feedback...  Ok, assuming the article is correct, how many shops have a 16 way NT machine.  Really.  My thought is that if they need that much horsepower, they should invest in someone who's done a lot of work on it.

>not too much nt bashing please :-).

 

I can't bash it too much, I use it every day I'm at the office.  I don't really dislike it.  My workstation's been pretty stable and hasn't required a lot of effort to maintain.  However, it's not under a real heavy stress, aside from having lots and lots of windows open as I work.  It still amazes me how hard it is to find some relatively simple things, squirreled away in the morass of menus (or buried in the undocumented registry).  What makes me wonder, though, is what someone's been smoking when they start talking about NT and enterprise level scalable hardware at the same time.  I don't care if it's got every OS internal feature to supposedly handle it or not, it's just not that stable.  I hear constant complaints about it from development and support groups where I work and I've seen enough evidence myself that'd make me lear away from it.  I could, on the other hand, imagine running Linux on an 8 way Xeon server.  Or maybe putting together a Beowolf cluster for a lot of processing bandwidth.  These things seem feasable to me.

Greg