= Pre-Pre-Announce: High Performance Parallel Smalltalk Achieved, Seek
Donating Subscribers
John B. Thiel
2026-june
Contents:
ABSTRACT
OVERVIEW
SEEK DONATING SUBSCRIBERS
FOOTNOTE: SPECIAL NOTE TO PROFESSORS / ACADEMICS
DISCLAIMER
COPYRIGHT
== ABSTRACT
A High Performance Multi-Core Parallel Smalltalk System is under
development and operational. I am seeking donating subscribers to
begin support now, to reward my work to date and encourage bringing the
system into existence and availability. Subscribers *may* be eligible
for early access and news (no guarantees).
== OVERVIEW
The system expresses a multi-year design effort, has succeeded
operational bringup, and feature buildout is proceeding. It has
baseline performance comparable to Squeak/Cog, plus near linear Nx
speedups across N cores for suitable algorithms.
It is a well-grounded design, with reference to standards and
theoretical foundations, and selection of high quality substrates.
The system has been designed from day 1 for threadsafe, high
performance parallelism including such iconic capabilities as efficient
Threadsafe #Become, and is true multi-core -- a single process can use
all cores.
It is an entirely new design and re-imagining of a Smalltalk system.
Not VM/Bytecode-based, nor an engine for any existing system. The
system transpiles to an intermediate language/substrate, which is
itself a powerful dynamic language, and thence to Native Machine Code.
It is not Image-based in the classical sense of "snapshot the world".
Objects must be manually persisted, and code is maintained in external
files. I consider this a preferable model, to leverage the wide array
of tools, editors, VCS, etc.
The system does have a live running image, and classes, methods and
objects can be dynamically loaded, defined and re-defined on-the-fly.
As a fundamental design feature, there is a comprehensive portable
external object format, capable to store/load most objects and
arbitrary complex *graphs of objects*.
The overall vision is a SYNTHESIS of two platforms: Smalltalk pure OOP,
message-passing, class library and dynamism, coupled with the
performance, features, native types and theoretical foundation of the
substrate language.
By deep design, essentially all native types of the substrate are
visible to the Smalltalk layer as Immediates, *including the types the
system is built with*. And likewise, all Smalltalk objects are visible
and can be messaged from the substrate language, so the entire class
library may eventually become available to programmers in the latter.
The Smalltalk layer is intended as First Class Citizen, eventually
fully self-contained, and programmers will not have to know the
intermediate language at all. Knowing it will avail certain additional
capabilities like defining new primitives on-the-fly, and fine tuned
optimizing.
* * *
No release-able product is available at this time.
There are major components and features remaining to be built,
plus porting and development of class libraries.
I estimate a timeframe of perhaps a couple years, maybe less,
until an experimental release suitable for expert developers.
However, the system is running on the bench with already many
capabilities, including object model, dispatch, subclassing, core
classes, Immediates, method definition, parallel thread fork, etc.
There is a 50+ page Design Manual, and several 100s R&D notes.
Measured performance:
160M+ sends/sec (peak seen is 210M+ sends/sec)
actual dispatched sends (i.e. not inlined)
to a standard object instance
on the fastest code path (all cache hits)
on a 1.6 GHz base i5-8265U (3.9 GHz boost).
Optimizations like inlining, no lookup, fast accessors, etc, are designed
and can yield even higher effective throughput.
Plus the Nx parallel multiplier across N cores.
== SEEK DONATING SUBSCRIBERS
This system likely cannot be free or open-source, at least initially.
It's extremely advanced, solves design challenges that other top minds
in computing have not accomplished in 50 years, and has taken years of
intensive research, design and development, unfunded, and personally
funded and initiated.
Prior Smalltalk parallel efforts involve greatly reduced single-core
performance (e.g. RoarVM at 22%), or cumbersome multi-process
approaches.
Whereas this system is running at near 100% baseline on single core
(ref. Squeak/Cog), transpiles at a high language level, and ultimately
compiles to native machine code.
If you are interested in this technology, I request you become a
donating subscriber now, to reward my work to date, and encourage and
fund bringing the system into existence and availability.
I'm inclined to a subscriber model because I want you to have ongoing
interest, and I intend that if and when the technology is ready for
wider release, first consideration, news and availability will be
offered to long-term subscribers in good standing (and you will be
expected to continue funding under some subscription license terms
and/or royalties) (but no guarantees in any case). You can of course
alter or suspend your subscription at any time. Also I welcome and
request your funding even if you don't have a personal immediate use
for this system, or prefer a 1-time donation for now.
A suggested contribution is 1% of your "seat" for individuals, or
department headcount or project budget, with any amount appreciated.
Please contact me with your contribution schedule and preferred
channel, and indicate if you would like news updates.
Thank you!
I can accept most cryptocoins (preferred),
or Paypal, bank wire transfers, mailed paper checks,
or set up whatever service like Patreon / Ko-Fi
you are familiar with.
Note that you are making strictly a donation, gift, research and
development grant, regardless of amount. It is not a purchase or
investment, and does not entail any rights whatsoever -- not use,
access, support or any other return.
If you would like to do development work on this system, it's not ready
for that, even if you are willing to contribute for free. There may be
a tiered model in future with open-source class libraries you could
work on.
If your organization is interested in special license or acquiring the
whole IP outright, it's in the realm of possible. Contact me with
offers and inquiries, bearing in mind this is a high ehd design and
system, which I believe advances the state-of-art, and there are few or
zero precursors or comparable technologies available. A substantial
non-refundable payment will be required to even start talking on this
level, plus NDAs, so structure any dialog accordingly.
Thank you for your consideration,
and feel free to forward this message to other potential supporters.
This is JBThiel.
(Linux '92, and lots else).
-- JBThiel
jbthiel(a)gmail.com
jbthiel(a)pobox.com (Paypal)
Cryptocoin addresses avail on request.
== FOOTNOTE: SPECIAL NOTE TO PROFESSORS / ACADEMICS
If you are in academia, I request you to grant me some graduate
degrees from your related programs.
My work and this system solves challenges the industry at large has not
achieved in 50 years. There are a number of extremely complex issues
to making Smalltalk simultaneously multi-core, threadsafe, and
high-performance, readily PhD level theses and beyond, and I believe to
have identified and solved many or all of them with this design and
implementation.
For example, the esoteric #Become capability alone is a basket of
theses, extremely complex to achieve safely and efficiently under
multi-thread, non-stop-the-world conditions. It permeates and induces
requirements across the entire system and object model. The ANSI
Smalltalk Spec and prior technology does not even contemplate the
semantics of such situations, let alone solutions.
In this system I have worked out a meaningful semantics and robust,
high performance approach to such cases, and others, that does not
stop-world.
I probably don't have time to write conventional papers for your
department and publications. The substantiation of my theses is that no
generally viable solutions have been produced by other top researchers
and developers in the industry over prior decades, and this technology
is now designed, built, and working on my bench. This work may be
included by reference.
I have a strong academic background, and experienced and expert in
computing theory and practice, but don't have any doctorates yet
despite completing most of a graduate program, due to the buffets of
history and economy, plus the entire graduate institute I attended
folded before I finished, and it irks me.
I may actually make this a requirement for use by any professors or
academics, so saddle your posse and grant me some PhD's or similar
credential from your organization (can be honorary).
As further incentive, I may offer a free license to your institution,
or credits, promotion, or other benefits. Open to your suggestions and
collaboration.
Thank you for your consideration.
-- JBThiel
== DISCLAIMER
NOTE CAREFULLY: There is no release-able product currently. There may
never be. If there is, it may not be released to you, or at a suitable
price or terms or capability for your purposes or expectations,
regardless of your support or contribution level. The project could be
dropped, encounter unsolvable difficulties, be sold to private
investors, etc.
The foregoing presentation, while believed correct, may contain errors
in perspective, design, measurement, etc.
You are being asked to support this author, research and development,
with voluntary gifts, donations, research grants, with the sole
motivation of rewarding efforts to date and encourage bringing the
technology into existence, with no specific return to you of any kind.
Your contribution is *not an investment, not a purchase*. There is no
guarantee, rights, license or expectation of any product, results,
profits, usage, news, access, priority, support, or any other return or
interest of any kind for yourself, organization, or at large.
== COPYRIGHT
The foregoing material and announcement is Copyright 2026, JBThiel, all
rights reserved, and may not be copied, modified or adapted for any
purpose, except to support promotion and funding of this system and
author. The mentioned softwares and designs are wholly owned
confidential technologies of JBThiel.
== END, PRE-PRE-ANNOUNCE
