= 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 [email protected] [email protected] (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
