Re: native Oracle-port on Linux -- what would it take?

On Wed, 17 Dec 1997 14:41:34 -0500, D. Jeff Dionne <jeff_at_maribor.pfnet.com> posted:
>[chop]
>> > strange really; it has become a FAQ how to install SCO binaries on
>> > Linux and quite a few sources say that most large companies or
>> > companies that have anything to do with the Internet, use Linux --
>> > yet
>> > Oracle are unable to deliver a Linux version of their database.
>>
>> Linux is missing many features for any OS to be worth considering
>> as a viable database platform.
>
>Yet it runs just fine under iBCS emulation on Linux...
>
>> 1. Linux has no logical volumes layer.
>> 2. Linux has no transaction-oriented filesystem.
>
>So write support for these, however...
>It runs just fine under iBCS emulation on Linux...

Both are under construction...

>> 3. Linux has no support for raw devices - hence NO even remote
>> possibility
>> to run Oracle Parallel Server.
>
>Have no idea what you mean. raw devices work just fine on Linux,
>tar -xvf /dev/fd0 for instance? Works fine. If Oracle needs raw
>partitions, no problem.

Raw devices are getting to be less and less of an issue over time what with the increasing levels of buffering being provided at the hardware level.

Linux does *not* at this time provide "raw" devices in the sense that Oracle (and I believe Informix) expects; it provides "cooked" devices where cacheing and buffering is *guaranteed* to take place at the OS level.

Oracle Parallel Server is probably not the first product that would be likely to be ported to UNIX, and I've never worked with an Oracle installation yet that didn't use "cooked" disks, so the point is moot for a lot of even rather large installations...

>> 4. Linux networking is flaky at best.
>
>Then the whole internet is flakey, since Linux runs on more servers
>on the net than anything else. Rock solid is more like it.
>It runs just fine under iBCS emulation on Linux...

The BSDs have a longer history of high levels of stability, and probably still maintain superiority over Linux. The issue being brought up is not whether Linux has "reasonably stable" networking, but rather whether it is robust under extreme conditions. Linux is a whole lot more stable than it used to be; certainly moreso than some of the platforms that Oracle runs on (NT, anyone?), but there is still room for improvement.

>> 5. Linux OS block size is what? 512 bytes? It would make even MS
>> Access laugh and puke steam.
>
>You're thinking of (an old version of) SunOS :-)

Variable, configurable block sizes have been around for quite a long time now. The claim that block sizes are limited to 512 bytes, and that this would necessarily be a problem, is certainly a silly one.

>> 6. Linux SMP is rudimentary and flaky at best.
>
>It's actually not that bad in 2.1 I hear, and I fail to see how that
>affects the viability of Oracle on Linux.

It's a fair enough criticism for those that would want to build database servers with 1GB of RAM, 300GB of disk, and multiple SCSI controllers to handle huge loads. You probably want to have 4 processors on that kind of box...

>> 7. The same for multithreading.
>
>But it works fine under iBCS on Linux. Interesting. You're mistaken.

Give threading six months to solidify under the GLIBC "regime," and we can provide a somewhat different story...

>> 8. Oracle requires efficient IPC and record-locking mechanisms. Linux
>> cannot do
>> record-locking at all, last time I checked. So Oracle would have to
>> write its own
>> code for that. Why would anyone sane want that?
>
>But it works fine under iBCS on Linux. Interesting. You're mistaken.

It's more interesting to consider whether or not this is something that Oracle normally needs to implement on the various systems that it supports. Sybase had similar concerns; they then lost their heads over the issue that if they patched the kernel to provide the kind of locking semantics that they wanted, they would need to GPL the kernel patch, and they didn't want that.

It's fairly unlikely that there are any changes required here that could not be readily handled if the Linux community were provided some sort of specification as to what the required locking semantics would be.

>> In fact porting Oracle to Linux would require Oracle to write its own Linux
>> almost
>> from scratch.
>
>But it works fine under iBCS on Linux. Interesting. You're mistaken

People run "toy" installations of SCO/Oracle atop Linux. SCO is not one of the high volume sorts of platforms; as a result, it's unlikely that anyone is throwing the kinds of challenging database "traffic" at the servers that would tickle the problems that would make them choke.

>> ==> I don't see much (or rather _any) prospects for seeing Oracle on
>> Linux.
>
>That may be so, but it has nothing to do with weakness in Linux. I
>suspect it has to do with Oracle's relationship with the Big UN*X
>Vendors.

There *are* some weaknesses. If Oracle (or Sybase or Informix) were to suggest a list of the most critical things that would need to be patched to get Linux "up to scratch" as a supportable platform, and the above list of 8 items overstates what would realistically be needed, there's nothing that couldn't be fairly quickly provided.

Oracle doesn't have a Linux version because they don't care to have a Linux version. There aren't enough suited people in the Linux community to "schmooze," and that's what Oracle is *really* into. Oracle is a great sales organization that happens to have a fairly decent database product. Much as IBM was, in the '60s and '70s, the most powerful *sales* organization on Planet Earth. (The fact of selling computers being largely incidental.) Microsoft has the greatest legal manipulation and acquisitions group of any company in modern times, as well as some sharp marketers. (Informix, on the other hand, has what seems to be a better database product, but they seem to be having trouble selling their way out of a wet paper bag.)

How much any of this has to do with whether Linux does or does not have any particular features is pretty hard to grasp at.

-- 
Christopher B. Browne, cbbrowne_at_hex.net, chris_browne_at_sdt.com
PGP Fingerprint: 10 5A 20 3C 39 5A D3 12  D9 54 26 22 FF 1F E9 16
URL: <http://www.hex.net/~cbbrowne/>
Bill Gates to his broker: "You idiot, I said $150 million on **SNAPPLE**!!!"