Re: native Oracle-port on Linux -- why bother?

From: Jay Thorne <jay_at_result.com>
Date: 1997/12/22
Message-ID: <349ECF6F.1505F34D@result.com>

Thomas Schenk wrote:
>
> tip wrote:
> >
> > there is NO WAY the pc hardware end of linux would be able to handle running
> > oracle in any decent manner. plus its tweaking is rather limited compared to
> > solaris and hpux. i have run oracle on both suns and hp boxes - there's just
> > no way linux is ready (yet) for oracle.
> >
> > i currently run oracle 7.2.3 on hp9000/s800/k210's running hpux 10.10 with a
> > gig of memory each. that's alright for running oracle.
> >
> > but imagine a pc with linux and a motherboard limitation of 128M, a 200Mhz
> > processor... i don't think so...
> >
> > linux is good for being a workstation, or a small server - but the heavy duty
> > shit - better leave it to superior hardware and os's.
> >
>
> This is total crap. The entire Deja News service is run on Linux
> machines
> and Deja News has one of the largest databases on the Web. I also don't
> know what decade you are living in, but we have Linux SMP running on
> dual
> Pentium II, 300Mhz, with 512M and 28 Gigs of hard disk space, so that
> stuff
> about 128M limitation and 200Mhz is bogus. I am not saying that Linux
> is
> the end all and be all of operating systems, but don't write it off as
> inferior simply because it runs on PCs. Just because it isn't a PC
> doesn't
> make it superior hardware.
>

This is getting silly. The guy with the Gig of ram had to buy that much machine since oracle is so horribly slow. Its slow because of all the ridiculous things that that come with it. Transactions, rollbacks, logs, PL/SQL. Eeeew.

I've benchmarked other databases against Oracle, and its laughably slow. So why do you need a Gig of ram on a huge HP box? With a pair of PII's_at_300Mhz, 512MRam, Oracle will be reasonable speed, but a memory pig. Except for the largest transaction oriented banking systems, 512M is pretty useful. PII's at 300Mhz are faster than the big HP, benchmark wise. PCI is a bit slow for really huge memory & IO, but its within the right order of magnitude. The big SPARKs Alphas and HP boxes have twice or more the memory/IO bandwidth. You need that when your app starts to use a ton of RAM & Disk.

With Solaris on a Spark 20 with 256M of ram, one medium sized oracle web app grows to 120 M ram in 3 days. Puke. This is why the guy needs a Gig of RAM. I cross ported to another DB(Texpress) on Linux/x86 and the app is 5M after 30 days.

From my experience, Linux would get an extra 10 to 20% out of the hardware compared to any of the alternatives. Its easier to port to than SCO, tons faster than Solaris and SCO and NT, and way more reliable than NT or SCO. Solarix X86 we had no reliability issues with , nor with Linux, but we did with NT and SCO. My main linux server has been down 6 times in the last year, each of them for OS upgrades.

In a bank/high volume commercial/governmental situation, you need extremely reliable Hardware and Software. Thats what all the "Ridiculous" extra transaction features of Oracle are for. If I was suddenly a CIO at a bank, I would not consider running the credit card system on Linux. Not because of OS reliability, but because the stockholders would lynch me if my datacenter was not supported by a reliable company. The OS doesnt even enter into it. The hardware is the basic unit of reliability here. Oracle enters into it, only because you need serious hardware to get good speed because its so slow.

I'd be a dyed in the wool big iron guy (IBM,DEC,Sun,HP). Not only is the _MACHINE_ a different breed, (at $5Million for a 3090/600e or $400K for a really huge DEC Alpha with enough CPU/Disk/RAM/Support contract) but the _SUPPORT_ the vendor gives the bank is of a different breed. When you are an IBM, a Digital, or an HP or a Sun, you put a hardware/OS support guy on-site. No Linux vendors can do that, becase the profitability per shipped unit is not enough.

A $400,000 Dec box is quite a bit different of a machine than even a HUGE PC architecture server. Even compared to a $30K tricked out Intergraph or Compaq box. Big iron is more reliable, hardware wise. Its much more reliable software wise, because for $400K, you get the time to check the MTBF's properly, and regression test the OS on the actual hardware you ship in all sorts of horrible conditions. You test it with poor power and lots of electrical noise and radio noise around.

Until the NEED for a really huge centralized PC architecture box happens, there will not be a PC machine of the big iron reliability class either. There will be no need, because the prevailing wisdom is to put in another server, rather than buy a bigger single server. Its tremendously cheaper for most offices.

Now, if you need to have a huge central system (say you are Chase Manhattan) then you talk to vendors who make money from support, not from hardware/OS sales.

Said the Ex DEC Field Service engineer, who spent most of the Eighties in cold computer rooms with his hands in the guts of VAXes. (Where Field service was a profit center).

+-----------------------+-------------------------+------------------------+


> | Tom Schenk            | Use Linux!              | All opinions

-- 
Jay Thorne  The Net Result System Services jay_at_result.com   
http://net.result.com
Zoom 505 Effect page http://net.result.com/~jay 
Zoom 5xx series Patch Database: http://net.result.com/~jay/db.html

Received on Mon Dec 22 1997 - 00:00:00 CST