---

New SPECweb99 Benchmarks Shows The Scalability of Linux

By Marty Pitts, Linux
Today

In early July there was a lot of excitement about the SPECweb99
results
of two almost identical Dell machines, one running Tux
1.0 on Red Hat Linux and the other running Internet Information
Server 5.0 on Windows 2000. The results showed the Linux machine
with over 2.5 times the performance of the Windows 2000
machine.

Linux Today reader, Anthony
Awtrey
, noticed that recently another Dell machine has been
added to the SPECweb99 results page. Interestingly enough, this
machine has 8 processors.

Previous stories have focused on the comparision between the
results of Linux and Windows machines with 4 processors each. The
spotlight this round highlights the the scalability of Linux and
Tux 1.0 (a new Red Hat product) over 1, 2, 4 and 8 processors.

The Machines

  • The single processor machine is a PowerEdge 2400/667, running a
    667MHz Pentium IIIEB processor with 2GB of RAM, 256KB of secondary
    cache and 1 network controller, and had 6 Windows 2000 clients over
    a gigabit ethernet network requesting connections. It scored
    1270 on the SPECweb99 Benchmark.

    Full disclosure page
  • The dual processor machine is a PowerEdge 4400/800, running 2
    800MHz Pentium III Xeon processors with 4GB of RAM, 256KB of
    secondary cache and 2 network controllers, it had 12 Windows 2000
    clients over a gigabit ethernet network requesting connections. It
    scored 2200 on the SPECweb99 Benchmark.

    Full disclosure page
  • The quad processor machine is a PowerEdge 6400/700, running 4
    700MHz Pentium III Xeon processors with 8GB of RAM, 2MB of
    secondary cache and 4 network controllers, it had 20 Windows 2000
    clients over a gigabit ethernet network requesting connections. It
    scored 4200 on the SPECweb99 Benchmark.

    Full disclosure page
  • The eight processor machine is a PowerEdge 8450/700, running 8
    700MHz Pentium III Xeon processors with 32GB of RAM, 2MB of
    secondary cache and 8 network controllers, it had 40 Windows 2000
    clients over a gigabit ethernet network requesting connections. It
    scored 6387 on the SPECweb99 Bechmark.

    Full disclosure page

All of the client machines used were Dell Precision 410
workstations with duel 450 Pentium II processors, 128MB of RAM and
were running Windows 2000 Professional.

Some simple math shows the increase in performance as Tux 1.0
scales across processors. We see:

A 73.23% increase from a single proccessor system to a dual
system,
a 90.91% increase from a dual system to a quad system,
and a 52.07% increase from a quad system to a 8 processor
system.

You will notice on the disclosure pages the changes in hardware
from one machine to the next. This reflects what happens in the
real world. You would hardly order a single processor server to the
same hardware specifications as you would an eight processor
machine. The biggest changes show up in the different between the 2
and 4 processor servers. It should also be noted that beyond 4
CPUs, scaling becomes more difficult as the Intel bus architecture
only supports 4 CPUs per bus.

Moving to the Intel Xeon processors gives a hefty boost to the
performance when comparing a single CPU to a dual CPU machine. The
greatest jump in performance shows up in moving from a dual CPU
machine to a Quad CPU system. This seems contrary to what you would
expect, since the Quad system runs 700MHz Xeon processors, instead
of at 800MHz that the Dual CPU system uses. A source close to the
testing of the machines indicated that the difference in Secondary
Cache, 2MB as opposed to 256KB is the ‘critical
differentiator’.

I was able to contact a source who has knowledge about the
hardware used and the tests performed. He commented on the
differences in the hardware used:

The scaling that is shown by the Dell results
is somewhat inflated in the 1-2 and 2-4 comparisons due to the
changes in hardware. If you look closely, you see that the 1 CPU
result used a 667MHz PIII, while the 2 CPU result used 2 800MHz
Xeons, plus twice as much memory, and a more powerful memory
subsystem. Then, going from 2-4 processors, we went from 800MHz
Xeons to 700 MHz Xeons, which again is a step up in per-CPU power
(even though it doesn’t appear that way) because the cache size
went from 256K to 2MB.

What is TUX?

In a
previous article
Ingo Molnar, the author of TUX, said:

Tux is:

  • an event-based HTTP protocol stack providing encapsulated
    HTTP services to user-space code
  • object cache, where objects can be combined with dynamic
    content freely. You can witness this in the SPECweb99 submission,
    the SPECweb99 TUX dynamic module (every vendor has to write a
    SPECweb99 module to serve dynamic content) ’embedds’ TUX objects in
    dynamic replies. The TUX object cache is not a dumb in-memory
    HTTP-reply cache, it’s a complex object cache providing async disk
    IO capabilities to user-space as well. User-space application code
    can request objects from the cache and can use them in dynamic (or
    static) replies.
  • full fledged webserver providing HTTP 1.1 keepalive, CGI,
    logging, transparent redirection, and many other
    features.

So in our opinion TUX is a new and unique class of
webserver, there is no prior art implementing such kind of ‘HTTP
stack’ and ‘abstract object cache’ approach. It’s, I believe, a
completely new approach to webserving.

TUX 1.0 takes advantage of new features the 2.4 kernel including
the new Linux TCP/IP architecture, VFS cleanups and scalability
improvements, the per-CPU SLAB cache, the I/O scheduler and large
file support.

So when will TUX 1.0 be available? According to the ‘SPECweb99
Run and Reporting Rules General Availability Dates’:

All the system, hardware and software features
are required to be generally available on or before date of
publication, or within 3 months of the date of
publication.

As it turns out, Ingo Molnar has released alpha source code on
September 1st, (see related stories), with the 1.0 consumer release
due out by the end of September.

Ingo Molnar said in a previous interview (see related stories)
that the code will be released under the GPL. TUX will be
distributed via RPMs and Source Code. Since TUX is
kernel-subsystem, it will be vendor-neutral.

Benchmarks should not be your only consideration for hardware
and software decisions. Remember Mindcraft? These results, from a
company that sells both Windows 2000 and Linux machines, are an
indication of the bright future Linux has in the Enterprise. With
the coming release of the 2.4 Kernel, Linux has the advantage over
proprietary software.

For the average user, having a system capable of such SMP
performance may not seem to be much of an advantage. Most of us
don’t have the need for a 4 or 8 CPU system. But with the increase
in scalability soon to be available, one more excuse not to use
Linux in the Enterprise will be invalid. Increased use in the
Enterprise means more applications and a larger mind share.

For a certainty, we are entering the halcyon days of Linux.

Here more
information on the SPECweb99 Benchmark.