Re: Linux betas NT in TPC testing, running Oracle8

In article <TOYV2.17252$95.541737_at_news2.giganews.com>,   cbbrowne_at_hex.net wrote:
> On Wed, 28 Apr 1999 08:11:44 -0500, Nik Simpson <ndsimpso_at_ingr.com> wrote:
> >r.e.ballard_at_usa.net wrote:
> >> : http://rpmfind.net/veillard/oracle/
> >>
> >> : This now points to a legal notice pointing out that they can no
> >> : longer publish the results. Too bad nobody glammed on to the bottom
> >> : line numbers. One of the reasons that there is such strict control
> >> : of TPC numbers is that vendors don't want $/TPC benchmarks done on
> >> : system like FreeBSD or Linux.
> >
> >You clearly have no idea about TPC benchmarks. In order to meet the
> >requirements of TPC/C on say a QUAD XEON you would need something around 4GB
> >of memory, 120-130 high performance SCSI disks, a large scale network to do
> >the testing, 20-30 client machines etc, etc. TPC/D is even worse in terms of
> >resource requirements. Basically, the only people with the money to run
> >these benchmarks are the large vendors, that said, there is aboslutely
> >nothing to prevent Redhat or VA Research from running validated TPC
> >benchmarks and I wouldn't be surprised if they did. But fundamentally the
> >TPC suite is aimed high-end OLTP environments were money is no object, and
> >incidentally, the price of the OS is a tiny tiny part of the cost of the
> >benchmarked configuration, so don't expect using LINUX to make a big
> >difference to the $/transaction (number of transactions for that matter.)

Look at what Beowulf has done to the Floating Point benchmarks. Linux clusters have now resulted in systems that bump into the top 300 and possibly even the top 500 fastest systems. The cost for this performance is roughly 1/10th the price of a comparable Solaris Enterprise 10000 system. The Avalon system which was 64 Alpha boxes at $150K was pulling the same numbers an the E-10K at $1M. IBM recently strung together 16 4 processor Netra servers and came up with a benchmark in the top 20.

Multi-gig drives, a gig of RAM, and raid in software can turn incredibly cheap PCs in the $400-500 range into pretty powerful machines. When you don't have to pay $1000/client for software and $100/user for client access licenses, the price/performance of a Sybase/Linux system can get pretty cheap.

> Add to that "lack of clue" the ridiculous implication that MTS/MSMQ
> were somehow cloned from Linux-based software.

Before there was MSMQ, there was System V IPC Message Queues, which Linux had back in 1993, maybe before. The SysV IPC message queues were wired up to processes that hosted TCP/IP connections.

At Dow Jones, a rather substantial research program into publish/subscribe systems intended for it's DowVision news wires added some pretty sophisticated capabilities such as what we now know as persistant queues, publish/subscribe, and queue managers that manage multiple queues was created. IBM/Level 8 used some of this technology (which had been given to about 30 financial services corporations and their consultants) and added support for LU 6.2/APPC, and a number of other communications channels (the original DowVision toolkit used TCP/IP and X.25). There was a Linux version of this software as well.

> MSMQ bears a "more than marked resemblance" to IBM's MQSeries message
> queueing system, which, as far as I can tell, does not run on Linux.

MSMQ does look very much like IBM's MQSeries. Both were initially developed by Level 8. MSMQ has the usual "Microsoft Incompatibilities" guaranteed to be incompatible with the rest of MQ Series. Level 8 sells a gateway between the two, but it introduces a major performance hit.

IBM has a port for the MQ Client for Linux. It's unsupported software, but it is available.

MQ Series is fully supported on a number of UNIX variants, but MQ Queue managers have not been ported to Linux (yet).

I pointed out before that Linux lacks a really robust Transaction Monitor. This does make me wonder how the original poster was able to generate TPC-D benchmarks without one. There is ACE/TAO, but I'm not sure you could do TPC-D efficiently with it.

> (FalconMQ *does* have a version of their MQSeries "clone" that runs on
> Linux, but that is quite recent, postdates their port to SCO, and note
> that it is designed to "talk" to an MSMQ server. You can't run an MQ
> clone on Linux at present without having an NT box around as the
> "controller.")

Or any number of UNIX boxes including Solaris, AIX, or HP_UX.

> As far as I am aware, the only TP "middleware" available for Linux is:
> - Tecco
> - Tibco
> - Tengah (maybe)
> - OM3
> - Cloverleaf
> - Orbix

Orbix has been ported to Linux? - Nice to hear! Does this include OTM?

> - Falcon MQ

> *NONE* of which were available for Linux at the time MSMQ/MTS was
> released.

When - exactly - was MSMQ/TS released into GA?

This would imply a fully capable implementation of MTS along with fully integrated MSMQ.

> [Note that Tuxedo, continuing in the tradition of TP products starting
> with "T," is rumored to be available for Linux, and I hear rumor of
> formal announcements next month...]

It's about time! :-)

> --
> "Microsoft: The People who Brought the Y2K Bug into Software Titling"
> -- cbbrowne_at_hex.net
> cbbrowne_at_ntlug.org- <http://www.ntlug.org/~cbbrowne/tpmonitor.html>
>

--
Rex Ballard - Open Source Advocate, Internet Architect, MIS Director
http://www.open4success.com

-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/       Search, Read, Discuss, or Start Your Own