Nice article; I wish all of the Linux hardware companies well.
Linux needs to be available pre-installed on computers available off-the-shelf at retail outlets for it to start invading the home in large numbers; a properly configured Linux box is not any harder to use than a Windows box.
After all, how many of your non-techie friends/family instlled their own copy of Windows? (As an example, Caldera is easier to install than Win98SE; reviews at my site)
Recently I showed a Mandrake installation running KDE to a friend of mine (who is a true dyed in wool Windows NT supporter / advocate) and his initial comment was: "It's just like Windows" - he had never previuosly evem tried to use a Linux box.
Well, as I recently found out when I reviewed Windows 98 SE, Linux can be easier to install than Win98SE. If this trend continues, then as more software is written (and more graphical, easy to use, admin tools appear) Linux will start making a dent in the corporate desktop, and then the general population.
BSD might have performed better for this test; however now that the problem is known I am sure there is going to be a patch released shortly that will fix the problematic locking. I think these tests were very useful; they pointed out several places where Linux could be improved / tuned.
With all of the pre-release (disappointing) K7 benchmarks floating out there I can hardly wait to get my hands on a K7 to test it thoroughly... AMD, are you listening?
From the information released so far about its internal architecture it should be a scorcher, but only extensive testing will tell the real story.
Actually the package listing was mostly from their web site; the outside of the packaging does not give nearly that much information.
As for the review being bland/dry, you have to remember that the majority of CPUReview's readers are solidly in the Win95/98/NT camp; I was trying to present the review from their point of view.
I did present a simplified view; the article was meant as a review not as a beginner's guide or howto. You are right, some of my suggestions were a bit advanced for a newbie; I tried to tone down the complexity but apparently I did not tune it down enough. I did not follow the "Workstation" install because in the past I ran into numerous problems when I did not do a custom/full install. I agree, the upgrade process could be smoother; I usually re-install from scratch every year or so.
Tweaking httpd.conf fixed the loading problem... I was not expecting quite the load I got from being/.'d so I had logging (and a customized referral log) turned on, not enough spare servers started etc.
Tuning Apache a bit got the load back under control (0.3-1.9)
Since it was a tuning problem, does that mean I was Mindcrafted?:-) (that was meant to be a joke, for the humour impaired; they are re-doing the benchmarks in an "open" enviroment I will be quite interested in the outcome)
I was trying to keep it simple for people new to Linux; depending on who I listen to my review was either too Linux oriented or not Linux oriented enough...
I personally would prefer a small/boot, a fairly big/home, a big/usr, a big/usr/spool on a separate spindle, and at least a/opt or/var; but I felt it would be counter-productive to go into that much partitioning detail in what was supposed to be a review.
You wrote: "If you run a commercial UNIX, you can expect close to linear scaling with multiple processors as everything is done very fine grain."
I have to disagree with you; unless you are running code that fits almost completely into the L2 cache you will not get anywhere near linear scaling with ANY OS - I don't care which one (yes, huge L2's help a lot, but I'll bet Linux 2.2 would also fly on an 8-way Xeon with 2Mb L2 cache each).
You have to remember that with shared memory SMP the processors have to share access to the main memory; and while the memory bandwidth does not get reduced to 1/N (N=number of CPU's) it does approach 1/N more closely with every CPU added...
Depending on the locality of reference, ignoring kernel locks, a big (>= 512k) L2 cache will give you about 90%-95% cache hits for workstation loads, but for server code you spend more time shuffling data for I/O or doing database searches; hardly L2 friendly. I doubt that more than 70%-80% of memory accesses are served out of the L2 cache on a big server.
Processor contention for memory resources can cause a drastic slowdown; a friend of mine once saw a dual processor box perform at 60% (yes, slower!) the speed of a slower single processor box! (there is an article about that on my site)
I really wish companies would "compete" via innoviation not litigation. Sony should just build a better "mousetrap" (console) that a G3 Mac could not emulate... than people would have to get a G4 Mac:-)
Seriously, competition is GOOD, gets us better products at lower prices.
Litigation only makes the lawyers rich, and wastes funds that could be better spent on R&D for more better products.
Ok, this is ridiculous. I should patent "autonomous atmospheric ingestion follwed by O2 removal and CO2 expulsion by bipedal mamals" (breathing) and collect a royalty of $0.0001 per breath! Or get injunctions to stop people from utilizing my patented method!
Ok, using a DejaNews search I found glide.xxedgexx.com, so I know there is an experimental X11 server for the Banshee. I could not find a reference on that site to a working Glide library for Linux/Banshee... Can anyone point me in the right direction? I bought (and reviewed at my site - shameless plug) a Gigabyte Banshee board, and I'd like to make full use of it under Linux as well as Win95.
The delays in release of NT 5.0 (oops Windows 2000) implies that they encountered more problems than they expected. If they drop the requirement for compatibility with old, nasty, hardware-manipulating code they can get it out the door faster. They need to release Win2000 by 2000, or Linux will eat their server market share for breakfast by 2001. If they do release by 2000, it may take until 2002:-)
Hmm... another reason to build an SMP box
on
Parallel Mesa
·
· Score: 1
I've been thinking of building an SMP box, and parallel Mesa is sounds like another good reason... Anyone tried the SMP mod to celeron A's?
Server support is nice, but I'd like to see Linux support for all their nice (drool) graphics hardware... Ok, I am a sucker for 3D graphics, and while the Riva TNT, Voodoo (2 & banshee), G200 etc. have given PC's decent 3D graphics, you can never have too fast 3D.
But I won't know until I test them head-to-head, which I plan to do.
I think KNI can be useful for games and 3D rendering (it should speed up matrix math something wonderful) but it really has nothing to do with normal business use, especially servers. How often does your file server need to raytrace?
Don't forget, just like software companies tried to use hard drive serial numbers, volume numbers, ethernet card MAC's they will probably try to use CPU serial numbers - see http://www.cpureview.com/art_is.html for an article on how it could be used for copy protection.
All I can say it they do this, AMD & Cyrix & IDT & Rise stock (and later Transmeta) will be good buys:-)
Rob's got huge expenses - bandwidth costs a fair bit - I know, I run a number of sites (granted, with *much* less traffic than/.) and I have to run ads to pay for the telecomm costs.
Remember, unlike me, he is doing this full-time, and also has to eat, pay rent, etc.
Now if he would only post announcements for me even once in a while
Slashdot Mirror
I'm running a petition to ask Intel NOT to remove the SMP capability from Celerons... I can use more signatures :-)
4957 signatures to date.
I'll try a BP6 (hopefully in the near future); when I picked up the M750i the BP6 was not announced yet.
Nice article; I wish all of the Linux hardware companies well.
Linux needs to be available pre-installed on computers available off-the-shelf at retail outlets for it to start invading the home in large numbers; a properly configured Linux box is not any harder to use than a Windows box.
After all, how many of your non-techie friends/family instlled their own copy of Windows? (As an example, Caldera is easier to install than Win98SE; reviews at my site)
Recently I showed a Mandrake installation running KDE to a friend of mine (who is a true dyed in wool Windows NT supporter / advocate) and his initial comment was: "It's just like Windows" - he had never previuosly evem tried to use a Linux box.
Well, as I recently found out when I reviewed Windows 98 SE, Linux can be easier to install than Win98SE. If this trend continues, then as more software is written (and more graphical, easy to use, admin tools appear) Linux will start making a dent in the corporate desktop, and then the general population.
BSD might have performed better for this test; however now that the problem is known I am sure there is going to be a patch released shortly that will fix the problematic locking. I think these tests were very useful; they pointed out several places where Linux could be improved / tuned.
With all of the pre-release (disappointing) K7 benchmarks floating out there I can hardly wait to get my hands on a K7 to test it thoroughly... AMD, are you listening?
From the information released so far about its internal architecture it should be a scorcher, but only extensive testing will tell the real story.
Actually the package listing was mostly from their web site; the outside of the packaging does not give nearly that much information.
As for the review being bland/dry, you have to remember that the majority of CPUReview's readers are solidly in the Win95/98/NT camp; I was trying to present the review from their point of view.
I did present a simplified view; the article was meant as a review not as a beginner's guide or howto.
You are right, some of my suggestions were a bit advanced for a newbie; I tried to tone down the complexity but apparently I did not tune it down enough.
I did not follow the "Workstation" install because in the past I ran into numerous problems when I did not do a custom/full install.
I agree, the upgrade process could be smoother; I usually re-install from scratch every year or so.
Tweaking httpd.conf fixed the loading problem... I was not expecting quite the load I got from being /.'d so I had logging (and a customized referral log) turned on, not enough spare servers started etc.
:-)
Tuning Apache a bit got the load back under control (0.3-1.9)
Since it was a tuning problem, does that mean I was Mindcrafted?
(that was meant to be a joke, for the humour impaired; they are re-doing the benchmarks in an "open" enviroment I will be quite interested in the outcome)
I was trying to keep it simple for people new to Linux; depending on who I listen to my review was either too Linux oriented or not Linux oriented enough...
/boot, a fairly big /home, a big /usr, a big /usr/spool on a separate spindle, and at least a /opt or /var; but I felt it would be counter-productive to go into that much partitioning detail in what was supposed to be a review.
I personally would prefer a small
You wrote:
"If you run a commercial UNIX, you can expect close to linear scaling with multiple processors as everything is done very fine grain."
I have to disagree with you; unless you are running code that fits almost completely into the L2 cache you will not get anywhere near linear scaling with ANY OS - I don't care which one (yes, huge L2's help a lot, but I'll bet Linux 2.2 would also fly on an 8-way Xeon with 2Mb L2 cache each).
You have to remember that with shared memory SMP the processors have to share access to the main memory; and while the memory bandwidth does not get reduced to 1/N (N=number of CPU's) it does approach 1/N more closely with every CPU added...
Depending on the locality of reference, ignoring kernel locks, a big (>= 512k) L2 cache will give you about 90%-95% cache hits for workstation loads, but for server code you spend more time shuffling data for I/O or doing database searches; hardly L2 friendly. I doubt that more than 70%-80% of memory accesses are served out of the L2 cache on a big server.
Processor contention for memory resources can cause a drastic slowdown; a friend of mine once saw a dual processor box perform at 60% (yes, slower!) the speed of a slower single processor box! (there is an article about that on my site)
I hope to review the K6-3 and P-3 at my site RSN.
For now, I've just posted a K6-2 400 review.
The K6-3 looks like a winner, I believe it is the first non-Alpha based micro with a three level cache architecture.
I really wish companies would "compete" via innoviation not litigation. Sony should just build a better "mousetrap" (console) that a G3 Mac could not emulate... than people would have to get a G4 Mac :-)
Seriously, competition is GOOD, gets us better products at lower prices.
Litigation only makes the lawyers rich, and wastes funds that could be better spent on R&D for more better products.
Ok, this is ridiculous. I should patent "autonomous atmospheric ingestion follwed by O2 removal and CO2 expulsion by bipedal mamals" (breathing) and collect a royalty of $0.0001 per breath! Or get injunctions to stop people from utilizing my patented method!
Ok, using a DejaNews search I found glide.xxedgexx.com, so I know there is an experimental X11 server for the Banshee. I could not find a reference on that site to a working Glide library for Linux/Banshee... Can anyone point me in the right direction? I bought (and reviewed at my site - shameless plug) a Gigabyte Banshee board, and I'd like to make full use of it under Linux as well as Win95.
I suspect it was /., or the plug was pulled (again)
The delays in release of NT 5.0 (oops Windows 2000) implies that they encountered more problems than they expected. If they drop the requirement for compatibility with old, nasty, hardware-manipulating code they can get it out the door faster. They need to release Win2000 by 2000, or Linux will eat their server market share for breakfast by 2001. If they do release by 2000, it may take until 2002 :-)
I've been thinking of building an SMP box, and parallel Mesa is sounds like another good reason... Anyone tried the SMP mod to celeron A's?
Did I not read of $250 Samsung 533Mhz Alpha's here a while ago? Why does not some company make a *cheap* Alpha/Linux server?
Server support is nice, but I'd like to see Linux support for all their nice (drool) graphics hardware... Ok, I am a sucker for 3D graphics, and while the Riva TNT, Voodoo (2 & banshee), G200 etc. have given PC's decent 3D graphics, you can never have too fast 3D.
This will be a fun year... K6-3, Jalepeno, P3 and then later, Transmeta and now this chip from TeraGen...
But memory sub-systems still need to be improved - the new CPU's are all starving for memory bandwidth.
But I won't know until I test them head-to-head, which I plan to do.
I think KNI can be useful for games and 3D rendering (it should speed up matrix math something wonderful) but it really has nothing to do with normal business use, especially servers. How often does your file server need to raytrace?
Don't forget, just like software companies tried to use hard drive serial numbers, volume numbers, ethernet card MAC's they will probably try to use CPU serial numbers - see http://www.cpureview.com/art_is.html for an article on how it could be used for copy protection.
:-)
All I can say it they do this, AMD & Cyrix & IDT & Rise stock (and later Transmeta) will be good buys
Rob's got huge expenses - bandwidth costs a fair bit - I know, I run a number of sites (granted, with *much* less traffic than /.) and I have to run ads to pay for the telecomm costs.
Remember, unlike me, he is doing this full-time, and also has to eat, pay rent, etc.
Now if he would only post announcements for me even once in a while