This discussion has been archived.
No new comments can be posted.
PC Week Reviews 2.2
Related Links Top of the: day, week, month.
A morsel of genuine history is a thing so rare as to be always valuable. -- Thomas Jefferson
A morsel of genuine history is a thing so rare as to be always valuable. -- Thomas Jefferson
NT only thinks it can handle SMP (Score:1)
And that's on better hardware.
*shudder*
Config files == source? (Score:1)
I think that GUI-dependent types may indeed think that tweaking textual config files is actually tweaking source. By this definition every Unix sysadmin tweaks source every day! I guess stepping up from there to editing Makefiles makes one a hacker eh?
I'm running 2.2.0... (Score:1)
The only reason I'd upgrade to 2.2.1 at this point is to get around that core thing... not that big a deal as my machine is primarily a desktop.
Duh! zdnet, pcweek are owned (mostly) by Micro$oft (Score:1)
> We ALL know that the 'Linux 2.0 doesn't scale,
> but 2.2 is on par with NT4' shit is crap.
Well actually, no. The curves and benchmarks that Henry publishes in his article are *completely* correct. I know, as I helped him run the benchmark.
Personally, apart from the 'tweaking config files' as source code error (and I actually tweaked the rc.local to change the max open files, max inodes and bdflush parameters to make sure the machine would reboot with the correct settings, not the smb.conf file) I though the article was completely accurate and very positive, thanks Henry !
Now the interesting question is why the NT Netbench numbers on the same machine in the same network configuration were not published as well......
But that's another story (and should be covered more fully later
Regards,
Jeremy Allison,
Samba Team.
Me looking stupid. (Score:1)
Examples:
glibc (threads)
make -j (fork
The Gimp (plug-ins are separate processes)
graphical mp3 players (fork off the GUI, the player, and sometimes even the oscilloscope)
Compare to Photoshop on the Mac, the only application that supported *any* kind of extra processor usage for a long time.
Or, for that matter, NT, which has much the same problem, actually, apparently unless you configure it properly and say the magic words and stuff... (I have yet to see this)
'steep learning curve' (Score:1)
Should we bring up the fact that there are GIGANTIC VOLUMES written about the registry? I know i've seen several ~2k page books... Given the fact that my Perl book is less than 1k pages...and that's for a complete programming language...
We need to destroy this FUD tactic.
Initially good, then very bad (Score:1)
Then it goes on to say "the community developing Linux moves at a snail's pace". What? When? Where? Last thing I knew I was installing RedHat 5.1, and now I'm on the 2.2 kernel. When did any other OS vendor move that fast?
Finally, they end on "Linux is still in flux". Who writes this FUD? Linux is far far far from being "in flux". NT2000, now there's an OS that's in flux. Linux 2.1.x was in flux. 2.2 isn't by definition, barring bug fixes and minor improvements.
Disappointing, although as they say - all press is good press.
Matt.
--
Dear PC Week, We'd rather edit conf files, idiots! (Score:1)
Matt - Win32:: module author.
--
Me looking stupid. (Score:1)
dearth of applications that can exploit
new SMP capabilities.
It's under the Linux Negitive side at the bottom of the page. Is there some exploit that screws up SMP Linux boxes running 2.2.0 or something? Or, is this his flowery way of saying that there aren't many things that really use SMP yet?
Believing it's the latter, isn't that totall BS because
Yes, not my subject, I could be totally wrong, but, I didn't think that that statement was totally true.
Funny lines in the article (Score:1)
Mainstream press attention void. (Score:1)
Whereas, there's a groundbreaking new release of Linux, and hardly a peep. Meanwhile, Linux in the press in general is healthy, but nothing concentrating on the release. I wonder if it's because of incoherent media co-ordination inherent to the open-source model? Any ideas?
--
Most Applications Already Parallel (Score:1)
You gotta use threads or separate processes and that requires a whole different approach than some applications use. You'd basically have to rewrite the code to use parallel algorithms.
Not true. There's an automatic speed increase when more than one job is running, since there's more CPU time to go around.
Note that netscape or quake are not threaded, therefore you wouldn't get get any performance gain with the exception that you could be using the other processor for other processes.
Netscape is threaded. Also, the X server runs in a separate process and can be scheduled on the other CPU, so that UI operations can be run asynchronously on another CPU. While Quake isn't multithreaded, the extra CPU keeps routine system operations (like email and web server) from slowing down your game. SMP makes for a massive improvement when you're playing quakeworld on the same machine that runs the server, since the server runs in a separate process and can be scheduled on the other CPU.
This might be good for something like ftp or http, where processes are forked off the main process (or inetd or whatever), but not for something like an SQL database.
It makes for a great improvement with HTTP servers, especially when lots of CGI scripts get running. SQL databases like PostgreSQL fork separate processes for each client connected, and I belive mySQL uses threads. Both SQL servers can easily be scheduled across multiple CPU's to decrease execution time when multiple queries are run.
All this talk about smp starts me drooling for an smp motherboard to learn more about all this stuff :)
Intel pr440fx, $100 on most discount hardware and auction sites. Has two PPro CPU sockets, 4 dimm sockets (up to 512mb), USB, 4 PCI slots, onboard UW SCSI, 2 IDE ports, onboard sound, onboard temperature, voltage, and fan speed monitoring, and 100baseTX ethernet.
Only vi???? (Score:1)
Of course I use vi, because every where I go it exists. (At least any machine I want to use.) It fits on a boot floppy, and once you know how to use it it's dam fast. You could say the same for emacs.
NT??? (Score:1)
Now that I read the whole article, they (PC Week) still missed the 'LINUX' point. Oh well...Maybe one day we can all live in a free software world....
Who cares which editor (Score:1)
covert Workstation to Server (Score:1)
Here, go to this link, and see the light, maybe even make a program that can bring others into the light.
http://www.uni-duesseldorf.de/ftp/pub/doc/books
program (Score:1)
Making sense of Makefiles (Score:1)
Good for you.
I tolerate all the config files with inconsistant syntax. If forced, I can even deal with fucking useless Scheme or LISP.
Personally I like LISP a lot, you should have a look at Haskell too.
Makefiles are another matter. I have a template one with explicit dependancies, so I normally just hack something out of that. For more advanced stuff though... Check this out:
Makefiles are very useful general purpose tool. Give functional (kinda LISP) and rule based (Makefiles) programming styles a fair go, it'll expand your mind much more than learning yet another imperative language will.
_Groovy_ horizontal rule.
modules: $(patsubst %, _mod_%, $(SUBDIRS))
$(patsubst %, _mod_%, $(SUBDIRS)) : include/linux/version.h
$(MAKE) -C $(patsubst _mod_%, %, $@) CFLAGS="$(CFLAGS) $(MODFLAGS)" MAKI NG_MODULES=1 modules
Let's not start with this one
What is that shit? Here is some more, from a different package:
%.o : %.c
$(strip $(CC) $(CFLAGS) -c $^)
Now %.o : %.c means this is a generic (also called implicit) rule for making any
The left hand %.o is the target and the right hand side %.c is the dependency. Make makes targets out of dependencies and saves (the computers) time by only making the target if it is older than the dependency (ie if the target is out of date).
Okay, consider $(strip $(CC) $(CFLAGS) -c $^) start from the inside and work out
$(CC) and $(CFLAGS) are environment variables defined previously (probably at the top of the Makefile).
$(CC) is your c compiler say egcs and $(CFLAGS) are optional flags to be passed to your compiler (for optimization, debugging etc), like I said it's optional so let's assume it's blank.
so $(strip $(CC) $(CFLAGS) -c $^) expands to $(strip egcs -c $^)
Now $^ is a special makefile variable (also known as an automatic variable). It evaluates to the list of dependencies, in this case there is just one %.c, so if foo.o was being made it would evaluate to foo.c.
So we are left with $(strip egcs -c foo.c)
Strip is a make function that strips out redundant white space so this evaluates to egcs -c foo.c
Cool, onto the second rule:
oldps w uptime tload free vmstat sessreg utmp: % : %.o
$(strip $(CC) $(LDFLAGS) -o $@ $ $(LIB_TGT) $(EXTRALIBS))
Now this is simple, it'll make any of the (multiple) targets (called %) by making the file %.o and executing the command
$(strip $(CC) $(LDFLAGS) -o $@ $where $@ is the target being made and $so executing make oldps will result in Make firing the second rule, with target oldps and dependecny oldps.o. This will result in the first rule being fired which will result in the command
egcs -c oldps.c
being executed if oldps.c is newer than oldps.o, now that the first rule has succeeded in making all its dependencies it executes
egcs -o oldps oldps.o
if oldps.o is newer than oldps.
That wasn't so hard was it? Now go change the makefile so it takes header (.h) files into account
A letter I wrote to the author... (Score:1)
While it is most appreciated that you are testing Linux esp. the new
kernel, a few point need to be made.
"The community developing for
Linux moves at a snail's pace."
Applications dealing with user friendliness have accelerated much
quicker than their Microsoft counterparts. Keep in mind that Linux is a
multi-user OS (as compared to Microsoft's single user OS, NT) and the
kernel will not be released until it is ready to be. Commercial
interests are nice on the desktop but they do not effect the development
of the kernel as the Linux community is under no commercial pressure to
release beta quality products under the guise of being enterprise ready.
Kernel development typically is much slower than application
development. Already both the KDE and GNOME desktops have more
functionality than the GUI of MS-Windows.
Installing Linux from a raw kernel is something that IS difficult. But
you can not classify installing a raw development kernel with a
commercial product release. I can install a Red Hat 5.2 server in 13
minutes and have it network ready with only one reboot. The same can not
be said for WindowsNT.
My 8 year old daughter installed Red Hat Linux at her school for science
day. No help from a grown-up needed. Red Hat 6.0 will have an even
easier install.
Again the point is do not paint Linux in a broad brush as Linux is a
comprehensive operating system with many facets.
"However, I
find it difficult to applaud a product
for mimicking the capabilities of its
competitors--"
This is an interesting comment as Linux has been 64 bit on the alpha
since 1995. Microsoft WindowsNT is still not at 64 bit. The clustering
ability in NT is little more than a fallback for overload and does not
compare with the PVM and SMP abilities of Linux. Clustering means
something different to a Linux engineer than to a Windows MCSE.
While it is true that may OS vendors have been providing SMP for a
while, hardware specifications had been closed up until recently making
development much slower than needed.
"Like Linux, Samba requires IT managers to dig into source code to
master it. We could reconfigure Samba 2.0 servers using the simple new
Web interface, but for more advanced configuration chores, such as the
reconfiguration of file cache memory, we had to use VI editor to
manually tweak our settings for optimal performance."
Any tweaks that are needed now will be fixed in the commercial releases
of Linux coming up. By the way the smb.conf file is not source code it
is simply an ascii text configuration file.
"Although administration lags, technical support option for Linux
are improving."
I'm not sure by what you mean by administration? Does Linux (as any good
OS) should have a technically competent individual to administer it.
LinuxConf (available at http://www.solucorp.qc.ca/linuxconf/) is a
fairly comprehensive administration solution allowing everything from
email,user,firewall,ftp,SAMABA,Apache,virtual domains, etc.
I haven't seen anything from Microsoft that comes close either on the
NT4.0 or the NT5.0 beta with it's management console.
Caldera has the LISA tool as well.
I commend you for your efforts at evaluating Linux however I would ask
you to evaluate all the possibilities available for it.
Good feedback on some of the technical aspects of your article can be
glanced at by viewing http://www.slashdot.org and look at the comments
section. Just a warning though...some people there can be rather harsh
so you may have to dig for the gold nuggets in the mudslinging.
Take Care,
Sincerely,
Nicholas Donovan
Linux Systems Group
Who cares which editor?? (Score:1)
Nick
LSG
(I feel like Rush L.) I Told You So! (Score:1)
After the trial they will say
"Linux is nothing compared to Windows2000"
and of course ZD-Net and it's ilk will fall in lock step.
Trust me. This is exactly how it will go down.
Nick
LSG
(I feel like Rush L.) I Told You So! (Score:1)
Give me one reason why I should trust you...
Registry (Score:1)
Well, if you'd look at those books, you'd discover that 90% of what's in there are the keys for your desktop wallpaper or screen saver or whatever. I'm sure if you published a book with every parameter used for the Linux kernal, KDE, WindowMaker, GNOME, Wine, X, sendmail, etc it would be 5 times as thick as those NT books.
Personally, I don't think the idea of a central configuation database is all that stupid. What's dumb is the structure that MS chose to implement, with application settings scattered to and fro in a manner such that not even the installers can keep track of it all. However, even in the mess that it's in, it's a lot more consistant than the myriad of unix config files with their widely differing structures.
Furthermore, the stated design of the Registry was that it was not supposed to be user-editable. Of course, there's a ton of useful settings in there that MS provided no dialog box for. For example, MS DNS (joke) virtually requires that you dig around in the reg. Furthermore, enforcing policies requires detailed knowledge of the reg.
(BTW, they are somewhat improving the sitation with a scripting interface that exists now in IIS and is system-wide in Win2K. I also should add that I've never seen an NT registry go corrupt like a 9x registry is certain to do.)
Config files == source? (Score:1)
(sorry bout the english)
ZDLabs (Score:1)
??????
be careful what you wish for... (Score:1)
But be careful what you wish for. Behind the buzz is the inevitable backlash when people discover that Linux is *not* like NT and that GUI-based program installation and system management is a bit funkier and requires a different mindset (thank Gopod for that).
However, as anyone who has wrestled with the update-and-rebind-and-reboot SEVEN times process when trying to work on TCP/IP configuration in Win 95/98/NT will attest, this is where Linux has it all over the Microsoft stuff. I installed DHCP on my Debian box, hooked it up to my DSL router and it just . . . worked. Not so with NT, which required a fair bit of finagling to figure out a very simple change I needed to make.
phred@sunlight.portland.or.us
-------
NT only thinks it can handle SMP (Score:1)
2.2.1 (Score:1)
Faster Web Servers (Score:1)
Why would Apache need to be changed?
Why wouldn't it be faster unchanged on 2.2?
Did someone tell ZDnet this? Or did they just "pluck it from the air"?
FUD, hidden in good press, abscured by compliments.. but, it's still FUD
ATI Framebuffer you lamer (Score:1)
Thanks
"Trouble is, just because it's obvious doesn't mean it's true"
Too bad they don't hilight ATYFB (Score:1)