From pyeatt@.cs.colostate.edu Wed May 7 15:38:46 PDT 1997
Article: 109121 of comp.os.linux.advocacy
Path: ix.netcom.com!howland.erols.net!newshub2.home.com!newshub1.home.com!news.home.com!nick.arc.nasa.gov!purdue!yuma!pyeatt
From: pyeatt@.cs.colostate.edu (Larry Pyeatt)
Newsgroups: comp.os.ms-windows.advocacy,comp.os.linux.advocacy
Subject: This is great!
Date: 7 May 1997 13:26:55 -0600
Organization: Colorado State University, Computer Science Department
Lines: 56
Sender: pyeatt@satchmo.cs.colostate.edu
Distribution: world
Message-ID: <5kql1v$rik@satchmo.cs.colostate.edu>
NNTP-Posting-Host: satchmo.cs.colostate.edu
Xref: ix.netcom.com comp.os.ms-windows.advocacy:199365 comp.os.linux.advocacy:109121
Yesterday I rebooted my computer. It was the first time in 2 months that I had shut down the system. Talk about stability!
I was working on a sound card driver and needed to run the Windows version under a debugger to see how it operates. Before shutting down, I converted an unneeded swap partition into a DOS partition. I got out a dusty box of floppies and installed DOS, then Windows, then the sound card drivers, then the video drivers. After only 4 hours, I had it running decently. I had forgotten what a pain it was to get a Windows system working. I was finally ready to install Borland C++ ( for the debugger ) I started through the install procedure and everything went smoothly until I got to floppy disk #22. Suddenly the computer claimed that there was some sort of terrible error writing to the hard disk. It tried again and again, but failed. OK, how do I get out of this? Oh, yeah, try Ctrl-Alt-Del. Oh! THAT's what they mean by the "blue screen of death." All this time I was thinking it was a cool FEATURE!
After rebooting, I tried the install again. Same results. The third time around, I only installed the debugger and nothing else. This time it succeeded. I am now 6 hours into the project. I ran the sound driver under the debugger. The system promplty crashed a dozen more times. After an hour, I found out that the debugger could not put breakpoints on IO ports without a mysterious driver file. Back to the install. Two hours later, after spending a total of 9 hours on this project and re-booting at least 50 times, I gave up and rebooted to Linux.
Guess what? Linux could not mount the root partition. OK. Boot floppy, Root floppy. Thanks YARD team! Check the root filesystem. It was hosed. fsck did the best it could, but basically, the directory structure was destroyed when Windows overwrote the inode tables during one of its crashes. Every bit of data on the hard disk ended up in lost+found. I just sat there, dumbfounded, wondering how anyone could buy anything as crappy as Windows. In a couple of minutes, the shock wore off and I started poking around in the lost+found. I decided that it was useless to try reconstructing the filesystems. It would take days. So I went back to /root and ran COMPLETERESTORE, a little script that I had included on the yard rescue set. One hour later, my system had been restored from the previous night's backup tape.
I mounted the dos partition on /mnt/crap and then typed "rm -rf /mnt/crap" I will never again use a MicroSoft product.
--
Larry D. Pyeatt All standard disclaimers apply.
pyeatt@cs.colostate.edu Void where prohibited.
http://www.cs.colostate.edu/~pyeatt
Richard Steiner wrote:
>
>
> (1) Microsoft OS products no longer "suck". Windows 3.1 did, but at
> least Windows 95 presents a semi-usable desktop and Windows NT is
> actually quite sophisticated in many ways.
Isn't it nice that after years of trying, microsoft has finally produced a "semi-usable desktop"? I work as a support engineer for a large company, where I support about 150 windows 95 users, and I'm here to tell you that windows 95 *does* suck. It leaks memory all over the place and is heinously unstable. Applications (including microsoft applications) gobble memory up and never give them back, even after the application is closed. I've verified this myself. Most of windows 95 features for idiots, which are supposed to be helpful, get in the way and cause more trouble. A concrete example: we recently had a server crash first thing in the morning at work. Users booted up their PCs and logged in. They go to run a program, from a shortcut, that resides on the crashed server. Windoze can't find the program, so it helpfully remaps the icon to a COMPLETELY UNRELATED APPLICATION because as far as it can figure, it's the closest thing to the program it can't find anymore.
So then I'm sitting there with 50 people with desktop icons pointing to the completely wrong application. Yay, microsoft! I won't even get into randomly disappearing printers, random spontaneous changes in network client settings, problems with microsofts tcp/ip implentation, or a host of other problems. We call microsoft, and all we get after a series of denials is "oh yah, that's a bug, we'll fix it one day."
Yay, microsoft! I'll admit, NT, except for the fact that it is a total pig, is much better than 95, but that's as far as I'll go.
> However, the majority of users puchase Microsoft OS products mainly
> because of one of more of the following:
>
> (a) It is all they know (or have heard of in the IT press).
> (b) It's what the applications they are familiar with need to run.
> (c) It's what others they know or work with use.
>
> Quality or functionality relative to other OS solutions is rarely
> an issue which is considered, since the MS OS product line is seen
> as being "good enough" for what is needed.
>
> (2) Linux has many applications. Windows has more applications because
> most of the larger ISVs are interested only in developing software
> for the one latform which maximizes their profit base.
>
> Division of programming resources between multiple platforms is
> seen as counter-productive, and many platforms which are more than
> deserving of support on purely technical grounds (Linux, OS/2) are
> left out of the picture based on market share, not capability.
>
> >Get off your hatered and jealosey! Linux is OLD news, just cos unix was
> >written before NT doesnt mean that its better.
>
> Be thankful that most larger computer shops still process most of their
> important data using the very machines you sneer at (UNIX servers, and
> large water-cooled pieces of Big Iron hosting disk farms).
>
> If the world was run with little toy boxes running wanna-be OSes like
> Windows NT (a good OS in its own context, but **NOT** God's Gift to mid-
> or large-scale dataservers as some claim!!), civilization as we know it
> would be hard-pressed to do much of anything.
> Here's a penny. Buy a clue, eh...
--
Rich Steiner >>>---> rsteiner@skypoint.com >>>--->
Bloomington, MN
Written online using slrn 0.9.3.2 and fte 0.45 for Linux!
The Theorem Theorem: If If, Then Then.
--
Brian Knox
bknox@richmond.infi.net
http://www.infi.net/~bknox
But wait, maybe linux can help. With a little savvy you can make a linux boot disk with a ramdisk containing a small set of useful utilities. You can boot into this micro-linux, mount the vfat partition, then simply do an "mv" command and put the windows directory back where it should be.
Another case where this micro-linux diskette would save win95's ass is if you had upgraded to a larger hard drive and wanted to just copy everything over from the old one to the new one. If you try this in DOS it won't work, the filenames will be lost. If you try it in win95 it won't work because there are too many protections + permissions--win95 won't let you mess with files that are of critical nature, such as the swap file. But if you boot into linux you can do a couple of mounts to mount the old and new hard drives, then do a "cp -a" to copy everything from one to the other! Then you remove the first hard drive from the system and boot off the new larger one, windows will come up and won't know it has been moved to a new hard drive.
The alternative is reinstalling windows then reinstalling every program you've ever installed. Or perhaps a lengthy backup on 1000+ floppy disks. Or just add the larger hard disk to the system but don't get rid of the old one. Too bad if you had plans for that old one in putting together a secondary system.
Space wrote:
>
> Say there are two Web servers. Both have exactly the same system
> specifications, a dual PII - 400 with 128MB of RAM and a T1 connection to
> the Internet.
>
> One of them is running Microsoft Windows NT Server 4.0 with all the Service
> Releases and 'hotfixes', and the latest version of MS Internet Information
> Server.
>
> The other is running Red Hat Linux, optimised for Web-serving, and Apache.
>
> After taking into the account the not-inconsiderable cost advantage to the
> Linux solution, option Two comes out on top. There is one question left:
> What would be the REAL WORLD performance difference (if any) between the two
> setups?
The only way you will know for sure is to try it yourself with your application, but here was our experience with identical Pentium Pro 200's
1. Performance:
In the beginning, NT appeared to be able to handle a few connections (1-5) faster than Linux, but after that, performance began to bog down and degrade while Linux remained consistent. After re-compiling our Linux kernel and Apache (using pentium pro optimizations) we saw substantial performance increase (~20%). NT cannot be recompiled -- it is optimized only with 486 flags.
NT performance would further degrade after several days. We were noticing that it was unnecessarily swapping. It appeared as if RAM was fragmented and NT could not efficiently allocate space for new client connections, so it was swapping out, or maybe there was a memory leak that was going uinreported. In any case, the only way we could solve the problem was to reboot about once a week. Linux has been running continously for almost a year without reboots (except for the one kernel recompile/upgrade).
Linux also serves as a DNS and mail server. NT was not able to handle those services well. DNS was difficult to configure and hung and sometimes crashed. The mail server (Exchange) was bloatware.
2. Stability:
NT has given us a few hangs -- where the machine locks up for no explained reason, and a few BSODs. Total, over the last year we have had to restart the machine probably 8-10 times for stability reasons (not counting the performance/memory issue we mentioned above).
Linux has only needed rebooting the one time after the kernel upgrade/recompile.
3. Administration:
NT administration was easier to get started. It was pre-installed on the machine, as was IIS. Once we got going, though, and wanted to make performance enhancements or administrative changes, things got more difficult. The IIS 3.0 interface did not give access to most server configuration settings (4.0 is better, but we have been finding more stability problems). To add a new host to the webserver we had to add a new IP and reboot the machine. To remove a host, we had to delete the IP and reboot the machine. Software upgrades required reboots. We had to constantly babysit the site and check Microsoft's FTP server for patches. Every patch meant another reboot. Most administration has to be done at the NT machine, which is inconvenient for the administrators.
File sharing has to be done using NT Appletalk services and NT Microsoft networking (SMB). Both (for obvious security and performance reasons) are limited to our LAN. The FTP services in NT were inadequate and inot secure enough for our needs (though there appears to be improvement with IIS 4.0).
Linux was harder to get started with. From the beginning, it appeared that our integrated netcontroller was not supported by the kernel. We added a supported NIC and were up and going after a little bit of struggle. The Linux and Apache configuration files were certainly not as friendly as the NT GUI administration tools, but there was a lot of power in being able to control the entire system using telnet and vi. We do not need to add new IPs to add new hosts, but when we do add new IPs we do not need to reboot the machine. Software upgrades are quite simple (we use RedHat). We have a compiler and a full set of useful libraries -- which makes programming much easier.
3. Security
NT seems to only have one kernel-level IP security option -- to turn selective IP ports on, or to turn all IP ports on. There is not option to filter hosts by mask or turn specific ports off. There is more security in IIS, but we found it similarly inadequate. For one, we cannot seem to specify security on a host level.
NT IIS does not seem to be able to be able to handle HTTP authentication permissions well for large user databases. Microsoft seems to want us to add a new NT user for each web user, which is totally ridiculous. We found a few 3rd party ISAPI modules which let you store user information in ODBC databases, but these were not flexible enough. NT authentication appears to be based solely on NT file permissions, which is ionconvenient if you want two web servers, with differen user bases, to share a directory.
NT has required rather constant security patches (downloaded from the MS FTP site). Each required taking the IIS server down and rebooting the machine. Many required that we also re-install IIS.
Linux had about half the number of security patches. Most were solved with configuration settings. Some were solved with software upgrades. Only one required re-comiling the kernel and rebooting (a process we integrated with upgrading and compiling our kenrel with better flags).
Conclusion:
We are encouraging our NT hosted clients to move to Linux as quickly as we can re-program their applications. This has meant more load on the Linux machine as clients are transitioning. Linux has been handling the additional load quite well.
We do have some clients dependent on NT -- clients using MS SQL or the like. It is proving expensive to move them to Linux.