Hacker News Reader

Ironically, I was later involved with the team at Oracle that converted Linux from a toy into a functioning kernel.

Seriously?

so full of himself...

In 1994 I did contract work for a serious businesses that ran commercial pay-for-click information website that ran fabulously on Linux, 1.2.x kernels.

Since I first started using Linux in 1993 (0.99 patch level 12), it has always been a functioning kernel.

Anyone can write a functioning kernel. The question is if it provides all the functionality required to run complex modern software -- especially big databases, which was pretty much your defacto benchmark back then.

"Linux doesn't provide these syscalls or anything comparable" might mean you go from "runs reasonably well" to "can't run on this hardware at all because the kernel doesn't know how to do these tricks with memory, etc"

I personally know at least a dozen Linux kernel contributors. Some of them are among the most competent, prolific and enthusiastic developers out there who would code until their fingertips bleed. Most work or worked at the usual suspects, yes, including Oracle.

It's the first time however I hear anyone claiming as much credit. Oracle didn't even start dabbling with Linux until 21st century, and it was already a decade old project dominating Internet server deployments at that point.

It certainly wasn't my intent to criticize any Linux contributor. I think that there are a lot of people doing the right things for the right reasons. They do that work in a social context, though. And I think that we need to understand how our society really works in order to make the right decisions moving forward.

Exactly. When I started doing the port of Oracle to Linux I had been reading the newsgroups - and they had been claiming POSIX compatibility. But system calls were missing many options, and entire sections of library code were missing. I'd run POSIX compatibility testing, so I knew that something didn't make sense. I kept posting to the news groups - mentioning that I was trying to port some software and hoping that they would notice my Oracle email address and take me seriously. Eventually, after trying several branches, somebody admitted that they had never actually run the POSIX suite, since it was too expensive. As long as you only wanted to do simple programs, Linux was fine. It was nowhere near being able to run Oracle, and it was nowhere near being able to run it well (with any of the then-modern features such as async I/O or direct memory paths or scheduling for long runnning processes or efficient large memory allocations.) It was a toy.

I'm with you on the facts, but it was anything but a toy. Minix was a toy, linux allowed a whole raft of useful applications to be run on PCs that would have otherwise required a large investment in hardware.

Linux would have succeeded with or without Oracle, and I find it sad that you would try to re-write history to the point where it would seem that without Oracle linux would have remained a toy.

> Linux would have succeeded with or without Oracle, and I find it sad that you would try to re-write history to the point where it would seem that without Oracle linux would have remained a toy.

Ehhh... I don't see where he claimed anything about that. I see "Linux was a toy. Oracle helped make it not a toy." But, nothing about what would have happened otherwise. Be careful not to get sad over things not said. Principle of Charity and all that...

Some circuit board with 64 kilobytes of RAM sent on a mission to Mars can be a "non toy".

Toy: for amusement or education of creator.

Non-toy: implements real requirement for some task.

For instance, 80 bytes of machine code which pop out a piece of bread when it toasts to a certain darkness level: non-toy.

>> I kept posting to the news groups

Which news groups were those? They (and your posts) should still be around somewhere.

I've found one post in comp.linux.os.development.

Exposing async I/O is largely a trivial hack, if the filesystem is already asynchronous under the hood, allowing I/O to proceed in parallel with computation. It's mostly about separating the request placement for I/O from being notified about completion or optionally waiting. The wherewithal to cob together this sort of mousetrap was in the Linux kernel since the early dawn, pretty much.

1990 POSIX didn't have aio_read and all that; it came later. A lot of this stuff was being standardized in parallel with the progress being made in Linux. According to the 1997 Single Unix Spec (Issue 2), the aio stuff came from the 1993 real-time extensions to POSIX. [http://pubs.opengroup.org/onlinepubs/007908799/xsh/aio_read....] "Derived from the POSIX Realtime Extension (1003.1b-1993/1003.1i-1995)". This was bleeding edge POSIX in the mid 1990's, and didn't exist when Torvalds started.

The performance optimization requirements of large databases aren't the be-all and end-all of "non-toyness", anyway.

The Linux people were decent at optimizing for the cases that their user base cared about. E.g. Linux exploded as a web server OS all over the world, so there was a lot of focus on decent networking, and tricks like serving a static file right out of the kernel and whatnot.

Thats pretty relative to be honest. I remember using a variety of Linuxes back then and being pretty unhappy with the results. Stability was barely a possibility. Compared to SunOS it was like Windows 3.x

I had a bunch of Linux servers in the mid 90's that we only rebooted once in 3 years, because of an office move. I'm sure there were ways of making Linux crash back then, but stability was easy as long as you got hardware that was known to work.

It's easy to forget the early days of Linux. If you go back and read the articles of the time they often pointed out that Linux made a good print server. But that was about all it could do - get lpr-style requests and talk to a printer driver. I did a test port of Oracle to Linux and wrote up a white paper for Larry. Larry was trying to figure out how to get Bill to kill important features in SQLServer. So Larry and Scott and IBM and HP had a couple of floors of kernel engineers at Redwood Shores fixing Linux and submitting the changes until Bill agreed to make SQL Server work more poorly stop competing so aggressively. But by that time we had already submitted enough to make Linux into a functioning product.

Curious what other things, aside from GNU and Linux, you fixed in passing on your corporate assignments.

I'm just a geek who grew up in Silicon Valley. I soldered and wire wrapped chips I picked up from Haltek/Halted and the sales offices of local chip manufacturers, and taught myself programming using switches to enter binary codes (before I bought a broken teletype and fixed it - paper tape was a huge improvement after using switches to program.)

I'm super curious about this. When was this, mid-90s?

well, it probably saved some lives by telling people that their printer was on fire

> Bill agreed to make SQL Server work more poorly stop competing so aggressively.

I might have been living under a rock and everybody but me already heard about this story. But I would expect that every HN reader would drop his coffee cup reading this for the first time.

Is this true? Why had I never heard about it? Why should Bill do that?

I'm not sure that I see the connection between Linux at Oracle and the group of "Larry, Scott, IBM and HP" wanting Bill Gates to neuter SqlServer. Could you explain in more detail? This whole saga is fascinating, especially since I've come from a generation where Linux has such a connotation of quality.

Yeah, I guess the connection isn't obvious. Bill had decided to start competing in the SQL space. He announced a bunch of features that matched Oracle's enterprise features 1:1 and promised to make SQLServer scalable and supported by a big consulting force. So Larry met with Scott/HP/IBM and to move forward with a "low cost linux appliance" with software to compete in the various segments of the PC space - they would almost give the whole thing away for free. Larry met with Bill with a list of demands in exchange for killing the project. It took a series of several meetings. I don't know the specifics asked by IBM/HP/Sun, since I only heard the Oracle side of the story. I believe we had, at one point, 200 engineers from those companies there at Oracle HQ making Linux work. I know that we flew in Linus to explain to him some of the changes (since they were enterprise kernel features that he had no background with) - and he green-lighted the changes (they really just matched what was already in the UNIX kernels from HP/IBM and others.)

I can remember those "Built a print server with Linux" times. But I did have the impression that MS really didn't take Linux as a serious contender, not even as a server OS. Was I wrong believing that?

I knew OS people from NeXT, Microsoft, and several other companies at the time. None of us took Linux too seriously - it was too unreliable and too immature. But by the time that Oracle got it fixed (and IBM "took over" Red Hat to roll out the "new" Linux), it was a real product.

There's a post on lwn.net from April 2, 2001 about the Linux 2.5 kernel summit:

https://lwn.net/2001/features/KernelSummit/

From that article: " The first presentation was by Lance Larsh of Oracle who, essentially, provided a laundry list of changes and features Oracle would like to see in order to get better performance out of high-end, large database servers "

...bullet list...

" The reception was "generally" positive - many or most of the above problems are well known and on the list to be fixed in 2.5 already. One exception was the non-preemption request, which was considered dangerous and unnecessary. "

J. Corbet is the author.

>> I've always found it ironic that the open source community ignores the role of the rich and powerful in allowing them to succeed.

I don't think anyone ignores anything, it's more like the OSS community (of which I'm not a part, btw) simply doesn't feel that their freedom is a good price to pay for - anything, really.

Also, to be fair, the "rich and powerful" mostly get into peoples' way when we try to live our lives in a meaningful way, and that includes OSS folk, who just like to write and exchange software. If the "rich and powerful" decide to patent everything, which is the real sticking point for the OSS community, then the OSS folk have no toys to play with.

Understandably, they are not happy with that.

I visited the NeXT factory in Fremont when it was making cubes. For fairly obvious reasons at UCB Sun workstations were everywhere. I was asked by some of the people giving the tour what it would take to get departments to go with their boxes. I said access to source was something we had come to expect but it was out of the question.

Yeah, that factory sure was impressive, wasn't it? I would sometimes go out of my way just to visit - hoping that they were running (it took a long time to get it started, and after that it took a long time to get it stable. Really, it was Sony and their manufacturing expertise that finally got us working.) I remember sitting up in the "control room" while they used the GUI to directly control the robots on the factory floor - watching the robots move in the dimly lit factory. It may be common now, but at the time that was "state of the art".

> I wrote the original white paper for Larry analyzing the state of the Linux kernel.

Would you mind sharing a link by any chance?

At the time it didn't seem like anything significant - nobody really wanted to do a Linux port of Oracle. But I did a test port and wrote up the many missing system calls and many incomplete system calls and many missing library implementations. Larry read it and I didn't really think about it again. Guess I should have saved a copy. And this was definitely an internal white paper - not something for circulation.

"open source" didn't exist during the time period you're talking about.

You sir, have some interesting stories to tell.

After browsing that user's comment history, it certainly seems like he's trying very hard to have some interesting stories to tell. Make of that what you will.

I second that.