Follow Slashdot stories on Twitter


Forgot your password?

IBM Exec Says no Large Web Servers on Linux 251

Accidental Angel writes "As reported in InfoWorld, Tony Occleshaw, IBM's software strategist for Europe, Middle East, and Africa said at CeBIT today that "No one runs large, million-hits-per-day Web sites on Linux." " Well, we served 640,000 pages on Wed on this Linux box. And the server load is only 2.00-3.00. I figure this box can handle around a million. The adfu server (also Linux) did around a million hits total that day, if you combine banner ads + layer HTML).
This discussion has been archived. No new comments can be posted.

IBM Exec Says no Large Web Servers on Linux

Comments Filter:
  • by Anonymous Coward
    If slashdot was an ecommerce site, they'd be able to afford a backup machine, along with other niceties that come along with making a box mission-critical.

    The crashes a few weeks ago were more or less because of administration mistakes.
  • by Anonymous Coward
    As David Filo, co-founder of Yahoo!, says in this [] article :
    FreeBSD has been extremely stable for us. We've seen over 180 days of uptime on a machine serving over 4 million HTTP requests per day. Performance has been impressive too. With disk striping using ccd we've been able to serve over 12 million HTTP requests per day on a PPro200 with 128MB of memory.

    You like cherries?
    How you like them cherries?!
  • by Anonymous Coward
    I run literally dozens of e-commerce site off solaris 2.5 and 2.6 using e450's up to an e10K
    for the portal, the stuff is solid and dependable
    I submit that configuration is ALL and can make or break a business, be it router/network config or local system config.
  • by Anonymous Coward
    As an archive site dejanews has lots of database servers too, all of which also run linux. I think their server farm is up to ~=200 machines now. I belive that they easily quailfy as 1,000,000 hit a day site. Ironically, if you take a look at their homepage, IBM appears to be one of their main sponsors.

  • "Maintaining a FreeBSD system is a real pain in the ass"? I beg to differ.

    I've run several FreeBSD systems in various roles from high traffic Samba, HTTP, and SMTP servers to the not so high traffic internal servers and even on a few desktops. I've _never_ had a problem with any of these systems.

    A FreeBSD install gets me a fully funtioning machine and source for all of Userland and obviously the kernel. I use CVSup to sync my source with the latest patches/updates/etc... everynight. Upgrades are as simple as 'make buildworld; shutdown now; make installworld; exit'. I've the latest system built from latest sources (That's Userland and KernelLand) - You can't do that with Linux, at least not with any distro I've ever seen.

    The ports collection makes installation of 3rd party software a breeze. The 'pkg' system on FreeBSD keeps track of dependencies, allows me to build from source, and always gives me a perfectly functioning binary in the end. The 'pkg' database is also _far less fragile_ than RPM.

    I ran Linux for years before I made the switch to *BSD. I left Linux because administration was such as pain. Since I've run FreeBSD I've been completely free of most of my administration tasks, these boxes "Just Work".
  • by Anonymous Coward
    Some months ago I was 'portunate' enough to be
    part of a team managing a major e=commerce
    venture running OpenMarkets Transact on solaris
    2.6. Admittedly the machine was initially poorely
    configured, but the thing was still seriously
    unstable, and it was running on two sun E3000s'.

    Frankly for mission critical stuff I would go for
    linux solely because I know the kind of probs it
    can have, and canget community help to solve them.
  • by Anonymous Coward
    I've been running a porn site for a while, and at times when I'm actively promoting the site, it get about 2 million hits a day. And that's a lowly Pentium 150 MHz, with 64MB RAM, old, slow, IDE disks, and 2.0.30-something kernel.

    It has NEVER crashed due to the web traffic. It has crashed a few times (4-5) during the last two years due to old, cheap and flaky hardware, but that's hardly something you can blame the OS for... (the little downtime there is doesn't warrant me buying new hardware yet)

    At the same time, the machine serve as a mailserver, and as my personal toy machine for misc. web projects that involve frequent gcc compiles etc.

    So Linux can't handle millions of hits a day?

    The problems Slashdot have had in the past, I suspect are more a result of extremely dynamical pages. Most sites stick with a bit more static content.

  • These are some stats (as of 1999-01-25) from one typical front-end (Linux 2.0.36/x86) to a rather large site:
    Roxen [] Challenger/1.2.34
    Booted on: Sat Nov 28 10:46:31 1998
    Time to boot: 1 sec
    Uptime: 58 days, 10:38:27

    Sent data: 130.26 GB 216.40 Kbit/sec
    Sent headers: 18.92 GB
    Number of requests: 100185573 1190.40/min
    Received data: 31.78 GB
    1190.4*60*24 == 1714176 hits/day

    Note that the above is the average over a ~2 month period (ie peek values are much higher).

    Of course, I wouldn't want to run the above server with Apache or similar. :-)

  • Apache? Not quite...
    internal-roxen-roxen [] gives a hint...
  • Must be your ISP. I have a T1 at work and a 56k modem at home ... either way I find that slashdot loads quickly, even though I am using a customized slashdot.

    I wish I had the chance to work on a million+ hit web site but so far i haven't been exposed to it. At the other end of the spectrum, I have setup a 386 with 8 MB of ram and a 90 MB drive ... it acts as a firewall for 50 PC clients and blows the doors of the Windows solution that was there before (both speed and reliablity). My stripped down Debian install uses only 33 MB of disk space! I would think that efficiency on low end hardware should translate into good performance on high-end equipment as well.

  • by jeremie ( 257 )
    I know it's commercial, but it's built on the same technologies(for the most part).

    I once administrated on a P90 with 32MB of ram running Apache 0.8.X, using BSDI, and it ran like a champ topping our T1 out at around 2 million hits/day(peaked at 4M/day frequently).

    There was quite a bit of kernel and apache tweaking involved, but it makes a statement for these technologies.
  • Nobody's going to bet their business on tuning a server to meet the needs of the clients? I can tell you one thing, if that business gets popular, then they're royally screwed.
  • 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)

  • IBM's not even on their own side. They're so big that their right hand doesn't know what their left hand is doing, much less how to give it a .. er.. hand. If you need more proof of this than the contradictory statements coming out of their executives, I have one word for you:


    Enough said?
  • No, that's neither a kernel nor a database problem. That's a filesystem problem, showing the limitations of the ext2fs filesystem. A database shouldn't have to implement its own filesystem just because the current one is extremely limited. On BeOS they don't have to either =)
  • FreeBSD != Linux.

    They have TONS of differences. Just because FreeBSD can do something, does not mean that Linux can do it (and vice versa). FreeBSD generally seems to be better at being a server. (witness's record-setting FreeBSD server).
  • I'd think that'd be a clue.
  • Posted by OGL:

    Have you considered your phone lines/isp? It's fine on this T3 connection.

  • Posted by FascDot Killed My Previous Use:

    I have to note that the two slowest things about /. are:

    1) is INCREDIBLY slow. I usually have to hit cancel to be able to see the page (which is fully loaded otherwise). This is unacceptable in a banner.

    2) sometimes does the same thing.

    AND, since I'm already on the topic of banners:

    /. Feature Idea: Since I have icons turned off, I'd like be able to move the banner to right next to the page title (assuming I have to see it at all). This would give me an extra inch at the top of the window allowing an extra story.
  • Posted by Nr9:

    the hotmail web page is run on free bsd while their mail servers are run on linux
  • by TopSpin ( 753 )
    Sun doesn't even make these any more. Are you sure you can
    qualify this as a 'major' e-commerce venture?
  • I'll take a good base and three live spares any day.
    ...when you read that as "three spare lives". Or maybe I'm just dyslexic :)
  • FreeBSD is no more reentrant in the kernel than Linux 2.2 is. About 96-97% last time I checked, which may as well be "not at all".

    Security is OpenBSD's forte.

    FreeBSD happened to be better able to handle brutal loads when Yahoo was setting up shop.

    "Nice" SMP is bullshit. Either kernel threads work or they don't. That's all there is to it.
    (well, as of 2.2; SMP on Linux 2.0 was poorer)

    If you want ultimate scalability and reliability you buy a mainframe. If not, you're taking your chances with hardware that is "only" singly or doubly redundant.

  • IBM ought to be in the business of pushing their mainframes for the real high end, where no one else has anywhere near their name recognition ("mommy, what's a Unisys?"..."isn't a Tandem a bike with two people on it?").

    This guy probably has to sell AIX (poor SOB) and doesn't want to admit that its JFS is the only good thing about AIX. (yes Virginia, DFS sucks)

    IBM ought to concentrate on
    1) mainframes in back, and
    2) Netfinity boxes running tuned Linux kernels in the middle;

    and let the desktop maroons choose their own OS.
    I.e. put Linux on a Thinkpad and sell it, now.
    Fuck, I'd buy one. (the VArBook 120 is $4000 and my old PowerPC notebook has booted its last)

    Stupid ass infighting. IBM is so close to (yet so far away from) the unity it needs to dominate without manipulating the market (ala Microsoft).

  • No, but you can give a "thank you Sybase" if you want to run a commercial RDBMS for free in deployment.

    Unlike Oracle, Sybase offers technically adept users the chance to run an older version for free, for full deployment (unsupported of course).

    Personally I don't see anything wrong with running the site on MySQL. Oracle is not bug free, Sybase is not bug free, DB/2 is not bug free, and Informix is not bug free. They're just big and baroque, and have been tested for longer.

  • First of all, it's actually Free. Period.

    Second, it handles transactions.

    Third, it is actively developed.

    Unfortunately, it has some shortcomings and is not "industrial-strength" in some areas. Plus it's wicked slow. (wait, a wicked slow RDBMS -- that's redundant!) However, for applications that are multiple-read, multiple-write, it is the only choice that's Free. MySQL is appropriate for multi-read, single-write, but not transactions at the enterprise (eg. bank, hospital) level.

  • A quick run of limit shows, and sets, per user, the maximum number of file handles one can have open.

    root@dogbert [~] cd /proc/sys/fs
    root@dogbert [fs] cat inode-max
    root@dogbert [fs] echo 10240 > inode-max
    root@dogbert [fs] cat inode-max
    root@dogbert [fs]

    Is that what you consider a "recompile?" Perhaps you're berating older versions of the kernel. Would you serve a million hits off SunOS 3? Is it relevant?
  • Slashdot had serious kernel level problems when it's hits started increasing. AFAIK CT had to do a recompile to fix it. Noone is going to bet a serious business on that.
    This is not "true." Slashdot falling over, blowing up, and becoming unresponsive is NOT Linux's fault. This is a webmaster's problem; this was Rob's problem. Proven by the hundreds or thousands of Linux machines that dutifully serve millions of hits per day, computer services offered are only as good as the system manager behind them. Rob is not perfect, Slashdot goes away some times; this is user error and can happen to any site on any network using any operating system.

    Linux and current hardware of the Intel IA32 or Alpha or Sparc generations have no problems saturating a 10 Mb connection for an entire day, week, or year. Of course, as you add in server side dynamic content, memory bandwidth and CPU requirements jump. It's the server manager's responsibility to make sure the hardware can meet the demands; if you need more CPU because you wrote scripts that require it, either tune your scripts or buy a faster CPU. Cluster your machines, do some DNS tricks, and farm out your web jobs. Perhaps invest in an SMP Alpha machine with plenty of I/O bandwidth an RAM. These are all server manager duties, and because so far the operating system has posed no problems, there is no call to "blame" the operating system for any of it.

    I would love to debate your argument on purely technical terms, instead of hypothetical duties and theoretical loads, but you don't seem to present any technical arguments against Linux. In fact, I would classify all your current arguments as FUD, plain Fear, Uncertainty, and Doubt.

    Sorry, I love Linux (and /.) but I wouldn't bet a million hits/day (average) web business on it. Imagine if Slashdot were an e-commerce site and had lost sales when it had it's "Problem week" a few weeks back.
    There's your standard FUD. "Oh no, there was a HUMAN ERROR in the past, what would happen if we extrapolated that to some fantasy future scenario and blamed it on the COMPUTER? Things would be mighty scary, indeed! Run from Linux, it is all things evil!" I'll tell you what would happen if Linux was serving an e-commerce site. Since this company has a brain or two, they have a backup web server, mirrored RAID systems, and do nightly backups. The main box goes down, the system administrators' beepers go off, they run into work, and see that the backup has automatically taken over web services for IP They get the original web server back on its feet, do some internal testing, sync the disk contents, and turn it back on. They go back home.

    Say the company wasn't so smart, didn't have secondary machines, and had to deal with not meeting performance. The system administrators, knowing the server scripts and hardware, would then look for things to optimize. Perhaps the machine needs more RAM, so it gets it, and they go home.

    If the machine isn't running in an hour, and the company loses a million dollars, then the system administrators are told they are now free to find other jobs.

    Tell me where, in those situations or others you may know, where Linux was a limiting factor here (I can only touch these hypotheticals because I have never seen Linux fail to serve content becuase it was suffering from any operating system problems). Mr. Sergeant don't include a single example of where a server didn't meet its expected load; I can only assume he's being compensated for his words by some third party. I would love to find out where he works, and what he's done with his life, but last I checked, his posted web address simply fetched a nicely printed Microsoft ODBC database connection error.

    I almost forgot to include your "inodes and shit" argument. Since no one here knows what you meant by that, and don't appear ready to explain, I can only guess that you are talking about maximum volume sizes of ext2 partitions on 32-bit machines. These are 32-bit architecture limitations, and here's a tip: get better hardware; buy an Alpha. Perhaps you want smaller inode _block_ sizes? Perhaps larger? Use a different filesystem, or at least give me a single, detailed, technical argument in favor of your sanity.

  • "for the European market, Linux is the No. 1 operating system used for Web sites" Isn't this true for every market? "For the RS/6000 AIX market, 0% of the servers run linux." I can understand IBM trying to sell AIX but sooner or later they'll realize the pointlessness of it.
  • Only on Sun boxes, but it is true.

    Plus, we have the (hands down) fastest network layer and process creation times.

  • the mail servers are Solaris

  • Linux handles 64-bit with the sole exception of physical RAM.
  • not constitute a Linux flaw on our part.

  • Now, let's count the total number of servers to power all of Over 200.
  • Really I've never heard of an NT server staying up during even 1000 hits an hour let alone an NT server running compilers and remote X clients during peak hours. When the corporate servers on my lan are crashing and burning the non commercial home Linux box keeps going. As long as VM doesn't run out.
  • What this means is that linux is taking root within IBM.. The second IBM starts to canabalize something it's in.

    IBM is way too big to focus 100% on something. OS/2 users know this first hand. The people within IBM who talk to the press are seldom technical people, they are seldom technically competant... This was one guy and linux probably competes with what he sells, he struggled and struggled to learn his product or get funding for it or something and now a free system is going to step on his turf? Not in his mind yet. This will get worse, IBM makes a lot of products that compete with linux and the people in those organizations are going to resist it. IBM also hires a huge number of "programmers" who only know Visual Studio and MFC, wait until you hear what they are going to say about linux when they start talking, it will boil your blood. This all just means that people are noticing linux and people are being expected to support it, it's really good news.

    In some respects his statments are correct, I don't think most thinkpad customers would be into linux at this point. Even with gnome and kde and staroffice and all the great stuff, it's still a little nerd oriented. On the server end, linux is awesome but it still hasn't proved itself under heavy fire, IBM ran the olympic site with absolutely huge amounts of traffic and it all kind of worked. Linux has run some big sites but not that big, yet. The people who pulled off the really big stuff will be the last to accept linux.

  • Everyone here is taking the approach that IBM is some sort of monolithic entity, as if one person at IBM being interviewed is speaking for IBM as a whole.

    I really doubt this is the case. IBM is, IMHO, more like a bunch of companies flying in loose formation. And fighting much of the time.

    If IBM had any unity at all, OS/2 might have beaten Windows on the desktop.

    This is just a way for one Pointy-Haired Manager to say something that serves two purposes:

    • For people who don't know much about linux, he's said something that makes it more likely that they'll avoid it, causing his division to prosper more than the division that was committing to linux.
    • Second, this will alienate linux users from buying from those same competing divisions of IBM that had committed to linux, thus making his division look even better by comparison.

    I doubt this person cares at all for how IBM does; he'll trade that away just to see his division out in front.

    His idea of "burying the competition" means taking market share (or simply destroying the market share) of a division down the hall.

    This is why everyone uses Windows instead of OS/2, and why, if IBM as a whole doesn't do something to deal with its unity problems, it'll wind up like Digital is now, IF they're lucky.
    Phil Fraering "Humans. Go Fig." - Rita

  • I think there are some scalability limitations with linux right now. The IBM dude was referring to (I hope for his credibility's sake anyway) milllionS of hits per type scalability. [I believe they use Alphas and Digital UNIX]..

    Of course, one would wonder how many people require that kind of scale, but businesspeople are (rightly, so) anticipating insane scalability requirements over the next decade as the herd "gets on the Net".

    While DejaNews does scale well, I think he's probably talking about more than database access here: we're talking commerce transactions and decision support systems (with complex querys).

    There are some things that are so large that you'd probably *want* to buy a Sun or Alpha box. Solaris is an excellently designed OS (and the source code is available to prove it)... Linux on those boxes probably would do a decent job too.

    Whether this will be true down the road (Linux 2.4/3.0) is another story.

    This also isn't just about scale, it's about reliability. Nothing is completely reliable of course, even goes down the odd time.. BUT.. Slashdot *HAS* had some noticable reliability problems in past (mostly due to MySQL if I recall correctly). Sun boxes are known for their reliability vs. Intel boxes where unless it's a 1st tier vendor you don't really know the quality.

    just my 2c

  • I wouldnt mind betting,, are getting those sorts of figures off their boxes.

    Are you sure that those would be anywhere near 1million hits/day? I mean, /. is "only" 600khd, and I can't imaging that they would be hit as much as /. is.

    Sure, they get a lot of ftp traffic too, but ftp traffic doesn't count toward hits/day. (Besides, ftp traffic is "easy" compared to most web traffic.)

  • Linux is seen by many people as a "light weight" server: good for small print spoolers, email systems, file servers, small web sites, small ISPs, low-performance routers, etc. They think that if you have a serious server, you need a serious OS.

    There is more than a grain of truth to this. Linux doesn't scale to a 64-processor, multi-GB RAM, multi-TB disk systems. It is just that these same people don't realize how much you can do with a 4-processor, 1GB RAM, 200GB disk Linux box, and how many PC users would think that is a large server.

  • Slashdot had serious kernel level problems when it's hits started increasing. AFAIK CT had to do a recompile to fix it. Noone is going to bet a serious business on that.

    Sorry, I love Linux (and /.) but I wouldn't bet a million hits/day (average) web business on it. Imagine if Slashdot were an e-commerce site and had lost sales when it had it's "Problem week" a few weeks back.
  • I helped Rob fix the inode problem (MAX_INODE in files.h) - the problem was that on hard hit sites MySQL hit the inode limit. (at least I think it's was inodes - that's why I said "and shit" - it might have been max-open-files).

    It required a kernel recompile and a reboot. So there.

    And to the other guy talking about static HTML - I'd trust Win95 or an Amiga to a static HTML site. That's not something to go by - any machine can serve 2 million static HTML pages a day.
  • IBM is positioning Linux as an NT alternative. It's only selling it in markets where NT is a serious contender. This is the low end of the server scale.

    IBM has 3 other high end server OSs. AIX, OS/400 & OS/390. I they put Linux in the same market they just fragment themselves more and loose money. Let them call Linux a low end server. It doesn't matter because people who deploy high end servers can dig through hype OK. If they couldn't NT would have supplanted AIX long ago.

    Linux doesn't need to be marketed on the high end. It just needs to be sold and supported. VA Research ships 8 CPU boxes next month. Linux care supports them immediately and IBM gets to discover high end Linux when it owns the market :)
  • is running
    Thisisarealoperatingsystemfromthefreeworld1.2alpha 12 on Linux

    Linux users include Slashdot,, and Deja News.

    It's probebly Appache. but boy did they customize :)
  • I don't think IBM is right, but I do understand how they came up with the statement. Managers understand surveys and rankings, its what they are trained to follow in the forms of trends and polls etc. What is needed to fight this FUD are concrete numbers from sources that management would find reliable.

    For instance Dataquest among others routinely publishes rankings of market penetration of various computer manufacturers products in a variety of different markets, or of operating systems into various broad application areas. From this a manager can read these published results and point at a source which states that the penatration of Windows NT into the million+ hit per day web server market is 99 and 44/100 percent or whatever the results say. These results are made from a poll of approximately 10000 companies.

    Here's where Linux runs into problems. Linux in general probably isn't even mentioned in the survey questionaires handed out to the various I/S managers who report these things to Dataquest. The sites most likely to be surveyed are also least likely to be polled by the Gartner Group, the people who run Dataquest. Slashdot for instance, while providing a valuable service to a rather large segment of the internet community doesn't register as a blip on their radar. It's just a few guys running a server as a hobby or if they're feeling kind a small business.

    Even if linux results were tallied there is a good chance that it would be further sub divided by distribution. So for instance, Red Hat would be a seperate operating system, Debian would be, Slackware would be etc. This might be fixable if these companies referred to themselves as value added distributors of Linux loudly enough that Linux is the product and RedHat, Debian et al become distributors.

    To help Linux be visible in polls it would be useful if a non profit company or co-operative were created which the various small businesses who rely on Linux could operate under. All of the information on servers, database size, web hits etc. would then be collated by this pseudo-company and delivered to the various fact finding organizations. Maybe small fees to be member (and to keep the company legally on the up and up) from which the entity can become an official client of one or more of the services.

  • I thought that DejaNews ran Linux.

    They must take millions of hits per day.

    Maybe the exec just doesnt know which sites run Linux because they dont get shouted about.
  • I prefer my Blue Axe of Learning[tm]
  • Transactions are a different beastie. And I can assure you that there are *VERY* few million transaction sites out there. Is this database transactions? If so, I suspect /. and others could and DO scale to this level nicely already. If you're talking financial transactions, then there are really NO sites like that on the Web. Why do I claim this? At a US dollar per transaction, a million transactions are a million dollars. Now, each of these transactions that happen, just so happen to be a lot more than that. Does anyone honestly think that any of these sites, like Amazon, rake in much more than $500-750k per day? I sure don't. Don't base opinions of how an e-commerce site will fare with what goes on here on /. First of all, changes would be thoroughly tested out on a staging server before even being deployed on an e-commerce site. Second, they'd be fielding a hell of a lot more hardware than Rob's tossing at it. Third, they would have more admin staff than /. has. Fourth, they'd be using someone like InterNAP to guarantee bandwidth- much of the slowdowns we see when the site's under load isn't in Rob's immediate pipe, it's the fact that he's on only ONE backbone (Amazon and many other high-volume e-commerce sites use InterNAP because they don't use their own backbone, they use the big-8's and route accordingly to avoid MAE-East and MAE-West 90-95% of the time.).
  • "p.s. how many of you -can- make a flaky Linux installation if you want to? I can -- compile the kernel with -O6, agressive strength-reduction etc."

    A trained monkey could do the same thing as you just described at any corporation you could name. Nobody in their right mind would do that- and that doesn't constitute "fickle" or capricious. To attribute what you just claimed to that is intellectual laziness on the part of the claimee.

    However, having said that, I will agree to some extent that this is all due to a percieved fickleness to the OS. Something we're going to have to work on for IBM.
  • If you deploy it in an application, it costs on a per user or per transaction basis. It would be prohibitively expensive for him to use it unless Larry (or one of his underlings) decided it would be good PR and an even better test of Oracle on Linux to have Rob use a "free" license.
  • By the very nature of IBM things are pulling in different directions. I'm sure those IBMers who released Jikes and several of my co-workers at IBM knows better. Even some 'suits' know better. But a vast majority don't and are trying to protect AIX. As someone else posted IBM is trying to position, even if it's not really true, as Linux as mid-range OS and AIX as high-range OS (Unixwise at least .. way too much NT use here, but that's another story). It's marketing, folks...
  • by kid ( 3373 )
    Take this from an IBM'er who previously has worked on the "outside" for MOST of his career;
    There is no "They" from IBM. You won't get a consistent opinion from everyone at IBM, merely different opinions from different heads of divisions. The clown who said this will be corrected, eventually. Assume the opinion comes from a much smaller company who has never dealt with the Americas. If you want to know how HUGE companies appear to the public, read this; lw-03-penguin.html
  • Ahem;
    Microchannel was the pre-cursor to today's video board design: Peripheral device intelligence reduces the load on the CPU. IDE vs. SCSI, no? It was a good idea.
    The only bad idea was that they tried to make money by licensing the design. VHS vs. Beta?
  • by luqin ( 3559 )
    actually, "maintaining" a FreeBSD system is a piece of cake. /usr/ports is a really nice way to install third party software. cvsup is a simple way to keep up to date with the latest source revision, and make buildworld/installworld is a piece of cake. If you think that FreeBSD is hard to maintain, then I'm guessing you refuse to read documentation. The current FreeBSD distribution system rules.

  • This is interesting could anyone expand on this??

    While not an expert here is my thought...

    I have heard that high end databases use there own filesystem for the database, and therefore don't have problems with clusters or inodes. On the other hand I think MySQL uses the linux filesystem for its database, and therefore is subject to the limits of the files system. So, rather than being a kernel problem this would actually be an issue of using a database not designed for the load being put on it.

  • Taken straight from the article: "We see Linux as a server phenomena right now more than as a desktop phenomena," and then several paragraphs later - "No one runs large, million-hits-per-day Web sites on Linux" So he thinks it's a server phenomena but can't handle a million hits per day? This doesn't make sense to me. It's like saying "IBM is an awesome company...nobody uses any IBM equipment or software for anything spectacular."
  • Stu Charlton wrote:
    I think there are some scalability limitations with linux right now.

    Out of curiosity... What do you see as being current scalability limitations? I'm quite curious.
  • I've got a few Linux boxes here getting 3M hits per day EACH!!! No problems...
  • To me, /. is perfectly fast. Sure, there have been a few reliability problems in recent weeks, *but* remember that /. was, up until very recently running on a dual PII/266 w/256MB RAM. And remember that /. is on a T3, (considering the number of cable modem and ADSL hits /. must get makes a T3 look pretty slow). All things considered, /. is not an eCommerce server. If it was, profits would mean a way faster server (probably AXP w/Linux), a backup system and more people to write scripts (and to debug them, etc.) Although Linux was origionally made to a be a PC UNIX, thus the lower end of UNIX, it's scaled quite nicely. I'd bet a multi-million a-day website running Linux, if I had like a quad Alpha 21264 with a good backup server. Although there are some scalability problems, Linux scales WAY more than SCO UnixWare or Windows NT (it's immeidate competitors, along with possibly Solaris/Intel). But I'd be happy if IBM would just include Linux as an option with all NT-class machines that it offers, as Linux is proven for these machines and is much more scaleable than NT is. Tim
  • That's 5,238 transactions a day.
  • You need to remember something though.

    /. != Linux.

    Yes Slashdot is running on Linux. But CT is always playing around with it. Also, as the kernels improve, so too should performance.

    Slashdot tested in on it's new box fairly solidly, and the errors received were more due to bugs in the code for the pages than from any load-up on the kernel.

    I propose that we attempt to Slashdot with all our might for a solid week. Just to test the stability of the box under heavy-duty loads.

    The only reasons the IBM people are saying "nobody runs it on million-hits-per-day servers" is because:

    1. Someone IS, and since there's no problems, they aren't griping about it.
    2. Nobody has actually attempted to test Linux under such a sustained load for long periods.

    {Pokes CT in the shoulder}, HEY ROB! YOU LISTENING?

    Chas - The one, the only.
    THANK GOD!!!

  • It sounds pretty cool - I wanna job there :-)
  • If you use a proxy you wont ever have to see adfu again (I don't). Rob isn't going to advertise this - he makes money off the ads, after all. You're expected to know better. For those that don't, well, deal with the ads.

    I haven't seen a adfu ad for months; /. is really fast for me, all the time. Well, during peak hours it might be a little slow.

  • I know Oracle sure as heck is NOT free for linux, unless you just want to download it and play around. That's about it. Has anyone used postgres for anything realy serious? It seems to be under very active development (they seem to have regular releases) and the developers sure do seem fired up about it :-) I haven't done much with it other than make a few tables ... is it really a suitable replacement for high end DB's? or maybe "middle end", where oracle would just be too much?
  • [Names changed to protect the innocent.]
    $ ll access_log.1 access_log.2
    -rw-r--r-- 1 root root 215507840 Mar 20 01:05 access_log.1
    -rw-r--r-- 1 root root 214266973 Mar 19 01:06 access_log.2
    $ wc -l access_log.1 ; wc -l access_log.2
    1083612 access_log.1
    1079518 access_log.2
    $ uname -a
    Linux server 2.0.36 #1 Tue Nov 24 05:29:56 PST 1998 i686

    And so, please pay no attention to the funny man with the "Thimk" license plate -- it is mearly more silliness from the people who brought you Microchannel(tm).

  • I dunno. I doubt a Pentium/90 sitting on a 128K ISDN line could serve millions of hits a day. But slap Apache in round-robin proxy mode in front of a cluster of Linux machines and you probably could.

    Right tool for the right job -- don't expect a smallass Pentium to go up against big iron, whatever the operating system. But provided you're willing to throw the right hardware at it, Linux can compete against the "big boys" just fine.
  • So, with "inodes and shit" you meant, that the standard kernel configuration as shipped by some distribution imposed a (let's say arbitrary) limit that could be changed easily and didn't even require down loading patches or actually fixing bugs ????

    Would this also mean that if Linus (or whoever worked on that piece of code) would have set a higher limit then you would consider linux suitable for a million pages a day ?

    Is it just me or do your statements just merely indicate that Rob and you spent a night hacking and cursing until you found why MySQL bombed and that you are pissed that it was just because of a silly, unnecessary limit set in the kernel?

    I'm glad you helped Rob getting Slashdot to work but please let your unpleasant experience not obscure the view on the whole issue. You can say you wouldn't run a high end web server on a plain vanilla RedHat/Debian/Suse/@&^%$&^% install, and everybody would agree. You could argue that there should be an easy to understand kernel configuration option "Configure for server [y/N]" that would raise some limits, and most of us would agree.

    The bottom line I seem to be getting here is that a lot of people got linux based high traffic web servers working after some tweaking, but there appear to be different opinions on how much tweaking is acceptable. This could be easily solved by releasing a distribution that is configured for high end servers (higher limits, some desktop stuff removed, basic server packages preconfigured and running, serial console, firewall, ...)
  • Her serveer is another Linux box which handles a huge (ludicrous even) number of hits.....
  • Just to get some more names, a french portal called voila [] runs Linux (look at the icon at the bottom of the page near the IExplore one). And so does DejaNews.

    But don't we just forget FreeBSD which has what might be the fastest tcp/ip stack for that kind of hardware. Just look at this nice bast*rd of who didn't let me in just because there were already 3600 users logged in. I guess this beast just explodes regularly most records and NT clusters Microsoft (r) can find.
  • What do you wanna run ? MacOSX ?

    sorry, just kidding.
    they get a million hits a day easy and they run
    redhat linux
  • My main gripe with slashdot's performance is the rendering speed. Massive tables just are dog slow. Not that I'm all that much better []. It does go to show that performance isn't just about cpu cycles. You can make a linux very slow [], and you can make NT scream [] (shudder), it's just easier to get good performance when you're working on something that isn't based on VMS. O'Reilly has a decient book on web performance tuning [] if you're intersting.
  • I do this every day. I run a site with millions of hits a day. It runs on Linux. It doesn't fall down -- day after day after day. High hit loads all the time. When we see 13 hits per second, we think that's low load. :^)

  • by xkahn ( 8544 ) on Saturday March 20, 1999 @07:07AM (#1971367) Homepage
    My company runs very large scale web sites on Linux right NOW. We
    get millions of hits a day, and Linux handles the load perfect, without
    complaint, and without problems or crashes. In fact, we use an Irix box
    for a file server on some of the sites and THAT machine crashes

    As far as we are concerned, Linux is the best investment our
    company has made. The machines it runs on are cheap and plentiful. The
    operating system is low cost and very fast and reliable.

    I think IBM is wrong when they say it shouldn't be used in these

    In addition, we also use Linux on the desktop for almost all
    employees. (There are a number of people in the company who need Word and
    so they don't use Linux.)
  • Hitachi? Don't they make gas grills?
  • by sled ( 10079 )
    They want to maintain the image that Linux is for "low end to mid-range" servers. That way they create a perceived demand for their proprietary OS in the "high end" server market.
  • Slashdot had serious kernel level problems when it's hits started increasing. AFAIK CT had to do a recompile to fix it. Noone is going to bet a serious business on that.

    Nobody is going to bet a serious business on an OS where they can tune the kernel to meet their particular needs that easily?

    You don't have the slightest clue what you're talking about.

    A serious business won't rely on an operating system where you *CAN'T* tune the kernel that way.

    I work for FedEx. We have recompiled kernels to meet specific needs thousands of times.

    Are you meaning to suggest that FedEx isn't a serious business? I bet we do more business every day than your company does in a year.
  • Scalability? Linux is more scalable than most OS's. my friends. We have Beowulf for PVM fun..
    and we have SMP, for 4-8+ way fun.

    Oooo, 8 whole processors!

    It gets awful after 4, sorry; circuit-switched SMP is like that, and always will be.

    Solaris scales well up to 64 processors right now, and the stuff in development is orders of magnitude beyond that. (Can't say more, under NDA, sorry.) HP-UX is the same way.

    Linux kicks ass on the low-end box, and when you put lots of low-end boxes together in a Beowulf cluster you can kick the pants off the big boys in terms of bang for the buck, but let's not do ourselves the disservice of calling it all things to all people on all platforms. That's Microsoft's style.
  • "does 64 processors" and "does 64 processors noticably faster than 8 processors" are two different statements.
  • Want to disagree with what I say on it? Fine.

    Here's what Linus Torvalds has to say on it, from his section in Open Sources:

    Symmetric Multi-Processing (SMP) is one area that will be developed. The 2.2 Linux kernel will handle four processors pretty well, and we'll develop it up to eight or sixteen processors. The support for more than four processors is already there, but not really. If you have more than four processors now, it's like throwing money at a dead horse. So that will certainly be improved.

    But, if people want sixty-four processors they'll have to use a special version of the kernel, because to put that support in the regular kernel would cause performance decreases for the normal users.

    In other words, Linus says EXACTLY the same thing I do! So grow up and stop doing Linux the disservice of claiming it's all things to all people. That's the tactic that's shooting Microsoft in the foot right now.

  • Well, there you go. Please do us Linux lovers a favor and forward this statement to the press and make that schmuck eat his words. I wouldn't know. I've never worked on IBM's unix, only Solaris, HPUX, SGI's, DEC's, and Linux, but this was obviously an intentional media slam on Linux in favor of IBM's unix.

    -El C.
  • OS/2 1.x was an example of too much "unity" over at IBM -- it ran well only on IBM MCA hardware, and included green screen emulators and other things that non-IBM shops did not want.

    I'm not sure if IBM really "unity", because they are just too damn big to sell an integrated solution like Microsoft does. If they did, it would be centered around the big iron mainframes, because that's where their money is. (And that is what they tried to do with OS/2 1.x.)

    Just the fact that they sell a Unix solution and are considering x86/PPC Linux prove that a little disunity is good over at IBM. (Although, let's face it, most of their revenue is from the installed base.)
  • FreeBSD doesn't NEED a distribution. The whole point is that BSD is scalable, meaning that you can run it on ANY kind of hardware without gratuitous apps eating up space. Coming from a BSD background, I can't understand what all you linux folks are going on about with distributions. What difference does it make? Why do you want preloaded software?

    Anyhoo - administration is simple for anyone with a moderate unix background. And, FreeBSD has been proven again and again to be able to handle higher loads with greater stability than Linux.
    17 linux servers running DejaNews? What does that prove? 17!??
  • The (lack of) speed of this site (if any) is due to bandwidth problems and pages being big and requiring the whole page to download before most (all?) browsers display them (because everything goes inside a big table). It has nothing to do with the site's operating system.

    It is very sad to see that many suits seem to think the way you do and take decitions based on irrelevant stuff.

  • According to netcraft [] hotmail is running on FreeBSD and Apache.

    And let us not forget: the godhead search-engine Google [] runs on Linux. I bet they get more than a lousy million hits a day.

  • Slashdot does 600,000 page views per day. HIts and page views are completely different stories (depending on how you count hits... and how you count pages, but I won't go there). The site I work for does 1.5 to 2 million hits per day (on Linux), but our page view count is much lower.

  • No, most major databases either require or have the option of using raw devices so that they can perform to the high demands of the system. Why put another layer between the database and disk if all of your multiple disks in your raid setup are used by the database? It's ignorant to do so.
  • It has already been mentioned that high-end production systems always should be customized. The max inode and/or max open files is just one example of such a configuration. Ever heard about configuring shared memory limits on a Solaris box for a database? I think changing those requires a reboot also. If so, what's you point?
  • I think, IBM is afraid of LINUX as a potential competitor for AIX.

    I agree with the posters who have already said that IBM probably doesn't give a damn about Linux competing with AIX. I believe that the reason they're offering Linux as an option is because of the unbelieveably huge amount of publicity and hype that has recently been thrown around the OS.

    (I personally believe that most of the publicity and hype is well-deserved, but that is of course just an opinion)
  • This statement shocks me. If they ran linux (which nmap confirms), why would they have such a crappy linux port of realplayer?

    Those people piss me off.
  • Yeah, and I have a wireless satellite-dish feed, which is technically a cable modem (if anything it adds a step or two) with 56k outgoing bandwidth. This adds a couple more steps and probably a whole new dimension of interference, yet the only problem I have is the adfu banner not loading, and thats no problem as I hate ads anyway. (and is NS 4.x specific?)

    And within the last week, I've read slashdot at each of the 24 hours a day contains.

    Seriously, check your own setup.
  • /. takes about a minute to load with Netscape 4.5 and it takes only a few seconds with Mozilla
  • The people at Realnetworks [] run linux
  • If you want an Alpha, buy an Alpha and put Linux on it. :-)

    About slashdot's stability, read rob's explanations of what happened. The instability usually falls under one of two categories:
    1. mySQL crashed from some bug in it.
    2. Rob saying, "noone should ever give me the root password to anything"

    Solution to 1: use a better database. mySQL isn't built for great scalability or much complexity, though the way that it is handling slashdot on not very much hardware is a decent testiment to it's scalability.

    Soltuion to 2: Don't give rob your root password. I.e., make sure to hire qualified administrators. This isn't a dig against rob, but I think that even he'd say that it isn't a fully qualified administrator for this kind of site, though has learned a great deal and is learning. (I'm not calling myself a qualified admin either, btw, so this isn't any sort of attack.)

  • by siberian ( 14177 ) on Saturday March 20, 1999 @11:35AM (#1971390)
    I dunno, I run 4 FreeBSD boxes and think its the greatest thing since sliced bread but Linux drives me insane and mystifies me at times.

    Its all in what your used to I think, not really an issue of true useability but rather human limitations in how we 'stick with what we know'
  • I don't think /. is the slowest site I visit. It's not *terribly* fast, but a lot of times it's faster than

    The argument about whether or not Linux can serve a million hits a day sort of misses the point: a million hits a day, in the grand scheme of things, doesn't require very high-end hardware. I'm pretty sure a two-processor Sun UE 450 could handle that with no problem, and that's a *workgroup* server. You'd need throughput of about 4MB/second for the peak, including processing....yeah. Should be doable.
  • I doubt it... Yahoo runs all its web servers on FreeBSD, so I'm certain someone's using Linux in the same type of environment.
  • by geophile ( 16995 ) <jao&geophile,com> on Saturday March 20, 1999 @07:27AM (#1971399) Homepage
    IBM has many parts, many of which are doing the same thing, few of which even know about the existence of these other groups. Furthermore, many of the IBM folks I've dealt with over the years have a severe case of NIH (Not Invented Here) syndrome. (If it's NIH it can't be good.)

    I can easily imagine an AS/400 manager thinking that Linux is his chance to make his machine even more relevant to the world. I can also imagine anyone who sells CICS or IMS or other highly proprietary software seeing Linux as just another enemy. (Someone in this position would be an ally of Java and EJB, however.)

    IBM will never be "on our side". Their interests may, in some situations, align with ours. Let's hope there are many such situations.

    I strongly recommend "Big Blues, The Unmaking of IBM" by Paul Carroll. It shows how a combination of arrogance, incompetence, and infighting led to the dominance of Microsoft at IBMs expense. You can't read this and not begin to see Microsoft emulating IBM. The book is also fascinating because it shows how a big, scary company can do such a consistent and thorough job of shooting itself in its foot over an extended period of time. Due, primarily, to "not getting it".

Trap full -- please empty.