The paper will explain how we did Simics-style checkpointing in SystemC, using the GreenSocs GreenConfig mechanisms to obtain an approximation for the Simics attribute system.
I while ago I wrote a blog post on checkpointing in virtual platforms, and what it is good for. Checkpointing has been a fairly rare feature in virtual platform tools for some reason, but it seems to be picking up some implementations. In particular, I recently noticed that Cadence added it to their simulator solutions a while ago (2007 according to their blog posts). There are a two blog posts by George Frazier of Cadence (“saving boot time” and “advanced usage“) that offer some insight into what is going on.
Virtutech and Cadence yesterday announced the integration of Virtutech Simics and Cadence ISX (Incisive Software Extensions), which is essentially a directed random test framework for software. With this tool integration, you can systematically test low-level software and the hardware-software (device driver) interface of a system, leveraging a virtual platform.
In my series (well, I have one previous post about checkpointing) about misunderstood simulation technology items, the turn has come to the most difficult of all it seems: determinism. Determinism is often misunderstood as meaning “unchanging” or “constant” behavior of the simulation. People tend to assume that a deterministic simulation will not reveal errors due to nondeterministic behavior or races in the modeled system, which is a complete misunderstanding. Determinism is a necessary feature of any simulation system that wants to be really helpful to its users, not an evil that hides errors.
One thing that surprises me is how rare the feature of checkpointing or snapshotting is in the land of virtual platforms, despite the obvious benefits of that feature. Indeed, checkpointing was one of the first cool things demonstrated to me when I joined Virtutech back in 2002. Today, I could not ever imagine doing without it. Not having checkpointing is like having a word processor where you only get to save once, when your document is finished, with no option of saving intermediate states.
But not everyone seems to consider this an important feature, judging from its relative rarity in the world of EDA and virtual platforms. Why is this? Let’s look at some possible explanations.
There is an eternal debate going on in virtual platform land over what the right kind of abstraction is for each job. Depending on background, people favor different levels. For those with a hardware background, more details tend to be the comfort zone, while for those with a software background like myself, we are quite comfortable with less details. I recently did some experiments about the use of quite low levels of hardware modeling details for early architecture exploration and system specification.
Unknown to most, IBM has one of the world’s longest records of using virtual platforms for software and firmware development and verification. This project has been ongoing since at least the days of the zSeries 900 machines, through z990, z9, and now z10. An excellent article on this virtual platform and its uses is found in the IBM Journal of Research and Development, number 1, 2009, . It is called “IBM System z10 Firmware Simulation”, by Körner et al.
Traditional hardware design languages like Verilog were designed to model naturally concurrent behavior, and they naturally leaned on a concept of threads to express this. This idea of independent threads was brought over into the design of SystemC, where it was manifested as cooperative multitasking using a user-level threading package. While threads might at first glance look “natural” as a modeling paradigm for hardware simulations, it is really not a good choice for high-performance simulation.
In practice, threading as a paradigm for software models of hardware circuits connected to a programmable processor brings more problems than it provides benefits in terms of “natural” modeling.
Now I am home again, and some days have passed since the IP 08 panel discussion about software and hardware virtual platforms. This was an EDA hardware-oriented conference, and thus the audience was quite interested in how to tie things to hardware design. Any case, it was a fun panel, and Pierre Bricaud did a good job of moderating and keeping things interesting.
I just read a fairly interesting book about the British Spitfire fighter plane of World War 2. The war bits were fairly boring, actually, but the development story was all the more interesting. I find it fascinating to read about how aviation engineers in the 1930s experiment and guess their way from the slow unwiedly biplanes of World War 1 and the 1920s to the sleek very fast aircraft of 1940 and beyond. It is a story that also has something tell us about contemporary software development and optimization.
I keep looking out for interesting examples of parallel software, and there is constant trickle of these. This past week I spotted a couple of new ones in the EDA field: SPICE simulation and chip timing analysis.
Cadence technical blogger Jason Andrews wrote a short piece a couple of days ago on his perception that host-based execution is becoming unncessary thanks to fast virtual platforms. In “Is Host-Code Execution History“, he tells the story of a technique from long time ago where a target program was executed directly on the host, and memory accesses captured and passed to a Verilog simulator. The problem being solved was the lack of a simulator for the MIPS processor in use, and the solution was pretty fast and easy to use. Quite interesting, and well worth a read.
However, like all host-compiled execution (which I also like to call API-level simulation) it suffered from some problems, and virtual platforms today might offer the speed of host-compiled simulation without all the problems.
Chip Design Magazine published an article by me in their August/September 2008, about Getting Software into the Hardware Design Loop. The article is about the technical and marketing aspects of how chip designers can get early feedback from software and systems designers, early in the hardware design process. The vehicle for this? Virtual platforms, obviously.
I just read the panel interview at the start of the latest issue (Number 4, 2008) of ACM Queue. Here, you have Bryan Cantrill of Sun (the man behind dTrace) bemoan the difficulty of testing faults. In particular:
Part of the reason I’m interested in virtualization is as a development methodology. It has not delivered on this, but one of the things that I ask is can I use virtualization to automate someone pulling the Ethernet cable out of the jack? I can get a lot closer to simulating it if you let me create a toy virtual machine than I can running on the live machine.
Well, this already exists. It is a common feature to any virtual platform that is not a datacenter-oriented runtime engine like VmWare, Xen, LPAR, and its ilk. Doing fault injection is a primary use case for virtual platforms, especially for larger servers and systems featuring redundancy and fault tolerance.
I just read an opinion-provoking piece “Software developer attitudes: just get on with it” by Frank Schirrmeister, as well as the article “Life imitating art: Hardware development imitating software development” by Glenn Perry that he linked to. Both these articles touch on the long-standing question of who does development the “best” in computing. I have heard these arguments many times, where software developers think that there is something mythical about hardware development that makes things work so much better with much fewer bugs, and hardware people looking at the speed of development and fanciful fireworks of coding that software engineers can do. It could be a case of the grass always looking greener on the other side… but there are some concrete things that are relevant here.
In the July 2008 issue of IEEE Computer, there is short article called “In Praise of Scripting: Real Programming Pragmatism“, by Ronald P. Loui, a professor at Washington University (WUSTL). The article deals with the issue of what is the appropriate first language to teach new CS (Computer Science) students, and considers that a “scripting” langauge like Python or Ruby might be way better than Java (no doubt about that I think).
What can this teach us for the purpose of simulation and the creation of models of computer system hardware for the purpose of simulation? Maybe a fair bit…
As might be evident from this blog, I do have a certain interest in history and the history of computing in particular. One aspect where computing and history collide in a not-so-nice way today is in the archiving of digital data for the long term. I just read an article at Forskning och Framsteg where they discuss some of the issues that use of digital computer systems and digital non-physical documents have on the long-term archival of our intellectual world of today. Basically, digital archives tend to rot in a variety of ways. I think virtual platform technology could play a role in preserving our digital heritage for the future.
The book “Taxonomies for the Development and Verification of Digital Systems“, edited by Brian Bailey, Grant Martin, and Thomas Andersson, was published in 2005 by Springer Verlag. It is a legacy of the defunct VSIA, and presents an attempt to bring order to nomenclature and taxonomies in the chip design field (its scope is defined to be broader than that, but in essence, the book is about SoC design for the most part).
In early July, Cadence announced their new “C2S” C-to-silicon compiler. This event was marked with some excitement and blogging in the EDA space (SCDSource, EDN-Wilson, CDM-Martin, to give some links for more reading). At core, I agree that what they are doing is fairly cool — taking an essentially hardware-unrelated sequential program in C and creating hardware from it. The kind of heavy technology that I have come to admire in the EDA space.
But I have to ask: why start with C?
In a funny coincidence, I published an article at SCDSource.com about the need for cycle-accurate models for virtual platforms on the same day that ARM announced that they were selling their cycle-accurate simulators and associated tool chain to Carbon Technology. That makes one wonder where cycle-accuracy is going, or whether it is a valid idea at all… is ARM right or am I right, or are we both right since we are talking about different things?
Let’s look at this in more detail.
I have another opinion piece published over at SCDsource.com. The title, “Why virtual platforms need cycle-accurate models“, was their creation, not mine, and I think it is a little bit off the main message of the piece.The follow-up discussion is also fairly interesting.
The key thing that I want to get across is that we need virtual platforms where we can spend most of our time executing in a fast, not-very-detailed mode to get the software somewhere interesting. Once we get to the interesting spot, we can then switch to more detailed models to get detailed information about the software behavior and especially its low-level timing. Getting to that point in detailed mode is impossible since it would take too much time.
This is something that computer architecture researchers have been doing for a very long time, just look at how toolsets like SimpleScalar and Simics with the Wisconsin GEMS system use fast mode for “positioning” and more detailed execution for “measurement”. It is also what is now commercial with the Simics Freescale QorIQ P4080 Hybrid virtual platform. Tensilica also have the ability to switch mode in their toolchain.
See an upcoming post for more on how to get at the cycle-accurate models – this was just to point out that that the article is there, for symmetry with previous posts about my articles popping up in places.
YouTube – Freescale QorIQ P4080 Hybrid Simulation is a video of a demo of the QorIQ P4080 hybrid simulation. Cool of Freescale to be publishing it like this, I think it is a very smart move!
Updated: Here is the video inline, let’s see if this works.
Only half an hour ago, the embargoes lifted. Freescale announced its new QorIQ series of multicore (and some single- and dual-core) processors. For the top-end of that line, the P4080, Freescale and Virtutech (where I work, remember) have developed a virtual platform solution to help Freescale customers get to working products faster. The virtual platform is available now, and is already running several operating systems including VxWorks, QNX, and a variety of Linuxes. Apart from the fairly large scale of this SoC, the really new part of the virtual platform is the so-called Hybrid solution, where the fast models are combined with detailed models from Freescale themselves. This creates a cycle-level detailed model with validated timing, “from the source” — but without the performance issues of having to run everything at great level of detail. Rather, you use the fast model to steer the simulation of a workload to an interesting spot, and then turn up the level of detail then and there. You can also select which components of the chip are actually detailed and which parts are modeled with the fast functional models, avoiding the incredible slow-down of running and entire virtual platform at a great level of detail.
If you happen to be at the FTF in Orlando, do come by and look at the demos!
I have been involved in this work for the past year, and it is wonderful to finally see it coming out and be able to talk about it.
As a follow-up to my previous post on the scope of ESL, I found a nice tidbit in an EETimes article… basically saying that hardware design is declining inside the typical system houses.
SystemC TLM-2.0 has just been released, and on the heels of that everyone in the EDA world is announcing various varieties of support. TLM-2.0-compliant models, tools that can run TLM-2.0 models, and existing modeling frameworks that are being updated to comply with the TLM-2.0 standard. All of this feeds a general feeling that the so-called Electronic System Level design market (according to Frank Schirrmeister of Synopsys, the term was coined by Gary Smith) is finally reaching a level of maturity where there is hope to grow the market by standards. This is something that has to happen, but it seems to be getting hijacked by a certain part of the market addressing the needs of a certain set of users.
There is more to virtual platforms than ESL. Much more. Remember the pure software people.
Edit: Maybe it is more correct to say “there is more to virtual platforms than SoC”, as that is what several very smart comments to this post has said. ESL is not necessarily tied to SoC, it is in theory at least a broader term. But currently, most tools retain an SoC focus.
Being a bit of a computer history buff, I am often struck by how most key concepts and ideas in computer science and computer architecture were all invented in some form or the other before 1970. And commonly by IBM. This goes for caches, virtual memory, pipelining, out-of-order execution, virtual machines, operating systems, multitasking, byte-code machines, etc. Even so, I have found a quite extraordinary example of this that actually surprised me in its range of modern techniques employed. This is a follow-up to a previous post, after having actually digested the paper I talked about earlier.
On Tuesday next week, I will be presenting at the Power Architecture Conference (PAC) in München, Germany. The topics will be multicore debug using virtual hardware, and the new Simics Accelerator technology. Especially Simics Accelerator is pretty interesting technology.
It is a simple idea, using multiple host cores to run a virtual platform, with fairly amazing results. Now, using a single computer we can run fairly incredible simulations that were the realm of pure fantasy just a few years ago. We also got a nice new little box to demonstrate it with, an eight-core Dell with 16 GB of RAM. With 64-bit Linux, this thing makes my Core 2 Duo laptop with 32-bit Vista look like yesteryear’s snail… And creates that giggling feeling that a really impressive new toy brings up in even the most grown up boys. Booting a 16-machine network of PowerPC boards was so fast it was not demoworthy. I think we have to up the ante to some 100 target machines to make it interesting, and I have no doubt that a combination of multithreading and idle-loop optimization will make that thing be usefully interactive from the target command lines. There are many other wild things we could try on that demo box, once it gets back from the Power Architecture Conferences tour.
By means of a trip down virtualization history, I found a real gem in 1969 paper called A program simulator by partial interpretation, by Kazuhiro Fuchi, Hozumi Tanaka, Yuriko Manago, Toshitsugu Yuba of the Japanese Government Electrotechnical Laboratory. It was published at the second symposium on Operating systems principles (SOSP) in 1969. It describes a system where regular target instructions are directly interpreted, and any privileged instructions are trapped and simulated. Very similar to how VmWare does it for x86, or any other modern virtualization solution.
Now the ESC SV 2008 is over. I really enjoyed going to the show this year, and presenting on simulation for embedded systems. The topic has to be heating up, I had some fifty people listen to the talk, which is really very good. Hope that they learnt how to build good transaction-level hardware models, and have some idea on how to apply this to their own projects. Hopefully, I can come back next year for the ESC 2009 (update: this did not happen) and do it again (even though the recent travel trouble makes it a less attractive idea to fly back here right now…).
It must have been Google Alerts that send me a link to the HOTOS 2007 (Hot Topics in Operating Systems) paper by Tal Garfinkel, Keith Adams, Andrew Warfield, and Jason Franklin called Compatibility is not Transparency: VMM Detection Myths and Realities. This paper is slightly less than a year old today, so it is old by blog standards and quite recent by research paper standards. It deals with the interesting problem of whether a virtual machine can be made undetectable by software running on it — and software that is trying to detect it. Their conclusion is that it is not feasible, and I agree with that. The reason WHY that is the case can use some more discussion, though… and here is my take on that issue from a Simics/embedded systems virtualization perspective.