Rick's b.log - entry 2016/09/26

Sorry for the delay...

We're not French or English, we're humans

Why Phoebe failed?

The RiscPC2 design was first thought about in 1996. The main stumbling block was the design of the IOMD2 which was being prototyped in an FPGA (programmable logic). The first working prototype with a real IOMD2 chip was built in 1998 just two days before the entire workstation division was terminated. It is unlikely any units would be shipped before 1999 at the earliest, had the project continued.
Costing an eye watering £1,499 excluding VAT for the base unit without any monitor, Acorn kit was always pricey but the truth lies in the fact that it just wasn't going to sell. Reports are that there were only around 200-300 preorders, a far cry from the hey-day of the BBC Micro.

The reason? It's actually quite clear now that time has provided us with a peek under the duvet. The Phoebe functional specification (a company confidential document) describes the RiscPC2 project.
Frankly, it sucks.

It would have been good, and it might have sold in 1996. By 1999, it was already outdated.

Here's the résumé:

The outline specification of the Phoebe project is described below.
 
Processor 233Mhz StrongARM on board running at 64MHz I/O
Performance 450k Dhrystones on a monitor supporting 70Hz modes.
Processor expansion card Interface to support upto 4 x 233Mhz StrongARMs on single processor card) . The card will not form part of this project.
FPA support not available
Floppy drive Single 2MB unformatted
Hard disc 1 T.B.D. size IDE HD with expansion for a further 3.
CD ROM/DVD ROM 32x slot CD ROM fitted as standard. DVD ROM upgrade possible
Total RAM SDRAM support only.
DRAM Expansion upto 512MB of SDRAM with 1 or 2, 32MB SDRAM DIMMs as standard.
ROM 4MB - board layout allocated for up to 8MB.
VRAM 4MB EDO VRAM fitted to board
Video up to 1024x768 non-interlaced in 32bpp. 1280x1024 24bpp
Drive bays 3 x 5.25", 1 x 3.5" visible bays, and 2 x 3.5" hidden bays
(1 x EIDE drive, CD and 3.5" floppy fitted as standard)
Free Drive Bays 2 x 5.25" visible expansion bays and 1 x 3.5" hidden as standard
PCI interface 4 slots compliant with PCI 2.1 spec (32Mhz 32Bit PCI)
Extended podule slots Backplane connector similar to RISC PC's allowing 3 slots
Case type and volume NLX Mini Tower case, volume = 450x450x190mm (approx.).
Network support 10 and 100 Base-T Ethernet support using a PCI card slot if req.
Parallel port Enhanced bi-directional multimode parallel port as per RISC PC
Serial ports 2 standard PC-AT style ports and IrDA support. Each will support baud rates up to 460kBaud (the actual performance achieved will depend on software running, user upgrades, screen mode etc.).
ID chip fitted as per RISC PC
Sound system 16 bit digital CD quality provided by Soundblaster compatible CODEC and VIDC20 sound systems. The Soundblaster sound and Acorn sound are mixed with CD/DVD output
Line input/output Stereo 3.5mm jack sockets on rear
Microphone input Mono 3.5mm jack socket on front of case
Internal speakers Mono
Headphone output Fitted on front, inserting headphone switches off speakers
Joystick 15 way D-type standard PC game port for dual joystick support
MIDI support MIDI input and output at rate of 31.25K baud on game port
PSU 230W NLX standard PSU

 
For comparison:
RISC PC700: ARM710, 50k Dhrystones in MODE 0, up to 2 IDE devices, 2MB floppy disc, 4,8 or 16MB RAM, 10,1,2MB VRAM, 800×600×32b video, 4 single Podule slots, Ethernet network card replacing one single Podule slot, bi-directional parallel port, serial port, headphone socket
The upgrade will offer in the region of 150-200% of the performance of a StrongARM RISC PC

Let's try a different comparison. If we're going to compare a premium with a premium, in January 1999 at the MacWorld Expo, Apple unveiled the new Power Mac G3 series. Costing from $1,599-3K (US$), you can get a 300-400MHz PowerPC G3 CPU on a 100MHz bus, up to 1GiB RAM, ATI Rage 128 graphics, 2×USB, 2×FireWire, 10/100 ethernet, etc etc.

Or to take it bit by bit:

Processor 233Mhz StrongARM on board running at 64MHz I/O
The problem here is a 66MHz bus was starting to get 'old' by '99; many computers shipped with it, but 100MHz was here. The bus speed relates to how fast data that isn't cached can be transferred between memory and the processor.
It is worth noting: Within this configuration, the ARM will almost always be running in 26-bit modes, as most of RISC OS (and all current user programs) will not run in 32-bit modes. So Acorn was still deferring the inevitable.

Processor expansion card Interface to support upto 4 x 233Mhz StrongARMs on single processor card) . The card will not form part of this project.
An interesting idea to have multiple processor cards, but would it work in practice? The entirety of RISC OS is based upon a single user single process methodology which makes me doubtful that RISC OS as it looks now would work effectively in a multiprocessor setup. We now have multicore processors (Pi2/Pi3 for instance) but RISC OS doesn't make use of them. I'm sure it will come, but in what sort of context, nobody knows yet. Remember, for what it is worth, that fair chunks of RISC OS (FileCore for instance) is not re-entrant. Remember also that pretty much all of RISC OS has no state memory other than "machine". What this means is that if I change directory in a TaskWindow, then this becomes the (new) current directory for the entire system. If I mess with the palette or the font quality settings, it doesn't affect just my application, it affects everything. The Desktop world is really a series of really clever hacks; but it is not enough of a multitasking environment to make the use of multiple processors (or cores) really feasible

FPA support not available
Yawn. Acorn never figured that people might want to make use of hardware FP mathematics. Why? Perhaps because the C compiler uses FP instructions for float (etc) and you'll be penalised for it. I mean really penalised. Why? Because on pretty much every RISC OS machine, all FPA instructions are emulated. Since the instruction does not exist, it raises the unknown instruction trap. The FPE module notices this, works out what the instruction wanted was, and then fakes the behaviour. Which means every RISC OS machine has access to a floating point co-processor. It's just that most of them aren't real.
To put this into context - I calculated a simple multiplication (123.456 × 654.321) many times. A hundred thousand? A million? I forget. At any rate, using the emulated FPA multiplication will complete the calculations in 388cs (that's nearly four seconds). Using the VFP (hardware) multiplication instruction completes the same sequence of calculations in seven centiseconds. That's a little over one twentieth of a second. This can have fundamental consequences to the behaviour of software that relies upon complex floating point calculations such as a ray tracer. Software emulation takes over fifty times more time than hardware. FP was standard on most PCs of the era.

DRAM Expansion upto 512MB of SDRAM with 1 or 2, 32MB SDRAM DIMMs as standard.
This isn't bad. While a PC of the era would struggle with 32MiB (I needed twice that in my PC to make W98SE comfortable), RISC OS has a much smaller memory requirement than, well, practically everything else in the world.

ROM 4MB - board layout allocated for up to 8MB.
While it is nice that the board can be configured for 8MiB ROMs (RISC OS is currently ~5MiB), wouldn't it have been more forward-thinking to have used FlashROM instead? The first CompactFlash memory cards were released in 1994, so this technology wasn't unknown...

VRAM 4MB EDO VRAM fitted to board
Video up to 1024x768 non-interlaced in 32bpp, 1280x1024 24bpp
And here we are seeing a distinct lack of foresight. 1024×768 wasn't an extraordinary screen size in the late '90s, it is about the largest that would cater for 32bpp. The display size that I use with my Pi (1280×1024) would be 24bpp - which probably wouldn't make any difference as RISC OS today still has rather rudimentary support for alpha transparency masks. This is, I should point out, space for a single frame of video data. By contrast, fairly basic PC video cards carried 8MiB onboard, and the decent ones carries 32-64MiB. Baking in a mere four with no upgrade path is surely built-in obsolescence?

Extended podule slots Backplane connector similar to RISC PC's allowing 3 slots
It is nice that the podule design would be continued in the RiscPC2, in addition to PCI, however is this going to be the same old boring podule interface, or will it have been extended to support the option of faster clocking?
You see, Acorn's built in IDE via the combi-chip was always really lacklustre (seriously, was it running in PIO mode or something? The IDE stuff inside my RiscPC got its ass handed to it by an ancient SCSI tape streamer!). This inherent naffness is part of the reason why people upgraded to third party podule-based filing systems. That, plus niceties such as partition support. Anyway, running flat out a podule is simply not going to be able to punt more than 6-7MiB/sec. Why? The fastest clock speed available is an 8MHz synchronous clock.
If the system bus can run at 64MHz, the podule interface ought to have "go fast" options as well.

Parallel port / Serial ports
There is a case for legacy hardware ports. The PC I'm writing this on has parallel, serial, PS/2 keyboard/mouse in addition to 2 USB on the front, and something like six around back.
But this is where Acorn could have stepped up and been a part of the game changers' club. They didn't need to take the Apple iMac route and toss all of the legacy ports in favour of USB; but for the RiscPC2 to have no USB (even in its v1.1 incarnation) is really rather woeful.

Each will support baud rates up to 460kBaud (the actual performance achieved will depend on software running, user upgrades, screen mode etc.).
Translation: We'll be providing a crappy serial implementation (that PC combi-I/O chip again?) that has lacklustre buffering, and we're still not going to fix DeviceFS. But don't complain, it's miles better than the really crappy half-hearted attempt at a serial port that was fitted into the Archimedes serial (and, I should point out, spawned a cottage industry in Acorn-wired serial cables).

ID chip fitted as per RISC PC
Because, yeah, using the network card's MAC address is too hard. Or, wait, was the network card going to be an option? In 1999? For real?

Phoebe, the RiscPC2, really would have been a nice upgrade from the RiscPC. Unfortunately, it took too long to get it to silicon, and by the time it arrived it was starting to show signs of obsolescence.
To put this into context, let's consider a different specification: An ARM processor clocking 600MHz, 128MiB of 200MHz DDR RAM, an NVIDIA GeForce graphics card with likely 32MiB onboard (can cope with 2048×1536 in 16M colours; and on a PCI card so upgrades might have been possible), UDMA100 IDE sockets, OS in FlashROM, Gigabit networking, USB2 ports. That's a real system. That's the Iyonix, first released in late 2002. You can see the difference.

In addition, the Iyonix marked the transition of RISC OS into the 32 bit realm. This is really something that Acorn should have introduced with the RiscPC, when the market was vibrant. Yet, we see with the information of the Phoebe proposals that Acorn were desperately clinging to the 26 bit realm, perhaps tied to the StrongARM processor as that was pretty much the fastest production processor that still supported 26 bit PC+PSR. Again, this was likely another nail in Phoebe's coffin. Why would somebody want to spend so much money in order to buy a moderately faster version of what they probably already have. I mean, okay, the system bus clocked 64MHz which is like five times faster than the aging hardware of the RiscPC; however the core processor was really not that exciting. A 233MHz StrongARM? I wonder how many people had that in their RiscPC already...

The weekend Orange died

On Saturday, blinking orange. Non-stop. So the Internet was no longer available, and neither was the telephone (as it was a VoIP phone).

Looking at Orange's suivi indidents web page (from my phone), it told me:

Sunday, half eight. Mom got up to look. I didn't bother. It's Orange. There's absolutely zero chance of it being fixed on a Sunday morning.
And, indeed, it wasn't.
Instead, the report page said this:

I loosely translate that to mean that they're not going to pay what their subcontractors want to fix a cock-up on a Sunday.

Also, I knew on Saturday something that it took them until Sunday to figure out:

So this morning, Monday morning, arrives. I know there will be no Internet, however the fault report said this:

It was working again by about quarter past eleven. Kind of makes my point about them not wanting to pay the techies to work on Sunday. ☺

Orange, as usual, have handled this absolutely abysmally.

• There is a mobile associated with my account. It's a Net+Phone+Mobile for which I pay nearly €70/month. At no time did they send me a text to say "we are aware of the problem". There has been no communication (text, email, anything) saying what happened to take out our internet for an entire weekend.
• You have the option to use your mobile phone as a modem, however data transferred will come out of your mobile data allowance. Sorry, Orange, but that's just not good enough. If you know there is a fault and it is your fault (not the client's), then the associated mobile phone should have unlimited data for the duration of the fault. This is for two reasons, firstly - modern PCs do a lot of crap behind the user's back (antivirus updates, for instance) which can eat a lot of data without anybody realising, and secondly - why should we be penalised for your problems? Unmetered access to data should be offered for the duration of the fault until the end of the day of the resolution of the problem.
• ...without any limits or throttling. If the person wants to spend the evening watching kittens in HD, why shouldn't they? They are currently paying you for a service you aren't delivering. And if this causes a load on the network and/or people to be abusive; well, maybe you should figure that into your cost analysis, rather than "we won't bother fixing it on Sunday 'cos it'll cost us".
• The help pages said that there would be an option in the "Orange et Moi" app to redirect calls to the landline to the mobile. I did not have any such option. Why? Is this a paid-for addition? Again, if there is a known fault of lengthy duration (more than an hour or two), this service should be offered. Yes, for free. It should be offered until the end of the day of the resolution of the problem.
• If I am paying sixty nine euros, if there are 30 days in this month, and if the service was down for a little over two days... that's a fiver for services not fully delivered. Do you plan any sort of restitution? Or even, if nothing else, some form of explanation/apology? You are supposed to be France's leading service provider. Try acting like it. If I wanted to be taken for a fool, I'd have stayed in the UK and signed up with TalkTalk.

I got by. I watched videos, I didn't do stuff that needed connectivity. I wrote the above text about Phoebe, because I knew I could upload it later on. I can't imagine things would have been so nice in a town house where entertainment relies upon the internet working (from TV to video on demand to Netflix), especially a house with children. And this isn't counting businesses that may have been using/relying upon the service working...

Your comments:

Please note that while I check this page every so often, I am not able to control what users write; therefore I disclaim all liability for unpleasant and/or infringing and/or defamatory material. Undesired content will be removed as soon as it is noticed. By leaving a comment, you agree not to post material that is illegal or in bad taste, and you should be aware that the time and your IP address are both recorded, should it be necessary to find out who you are. Oh, and don't bother trying to inline HTML. I'm not that stupid! ☺ ADDING COMMENTS DOES NOT WORK IF READING TRANSLATED VERSIONS.
 
You can now follow comment additions with the comment RSS feed. This is distinct from the b.log RSS feed, so you can subscribe to one or both as you wish.

VinceH, 28th September 2016, 22:45

"You are not reading my b.log using HTTPS. You can switch to HTTPS by clicking here."    RewriteEngine On  RewriteCond %{SERVER_PORT} 80  RewriteRule ^(.*)$ https://www.heyrick.co.uk/$1 [R=301,L]

VinceH, 28th September 2016, 22:50

Forgot to say, that should go in .htaccess - but you'd probably be able to work that bit out. :)

Rick, 28th September 2016, 23:15

Thanks for the code snippet. The latest entry (yup, you get two today!) explains why I'm specifically NOT transitioning everything to HTTPS.  Maybe in the future. But not today. I'll keep a note of the rewrite rules though, that's useful.

VinceH, 29th September 2016, 00:32

I've been doing it on softrock.co.uk for a while - I felt it necessary not because of Google (I fart in their general direction) but because I set up some client log-ins for some report data and forms for them to send stuff my way.    Annoyingly, the certificate is included FOC by my hosting company, but they only provide a one, irrespective of the number of domains - and I can't see a way to add more from third party sources*.    You can therefore visit (for example) RISCOSitory.com using https - but the certificate is for softrock.co.uk (and depending on your browser) you may be told it's invalid. :(    * Of course, that was when I set it up - I should check in case they've changed things.

Xavier Louis TARDY, 19th October 2019, 10:34

You write :  'Processor 233Mhz StrongARM on board running at 64MHz I/O  The problem here is a 66MHz bus was starting to get 'old' by '99; many computers shipped with it, but 100MHz was here'  but you seem to ignore the fact, per tick, ARM systems access more data than others.  How does 66 Mhz memory bus for the StrongARM CPU compare with 100 Mhz memory bus on the Intel CPUs available at the time ?  That's the only good reasoning, not a mere 100 > 66.

Rick, 19th October 2019, 11:29

Per tick, the ARM may appear to get more done, but it is worth remembering that the era we're talking about is the days of the early Pentium. A 32 bit processor with two internal 32 bit pipelines, and a 64 bit memory bus (remember the 32 bit SIMMs were supposed to be installed in matched pairs? that's why; newer machines used DIMMs).  By comparison, the StrongARM offered a single 32 bit data bus. So while simple bus speed comparisons don't tell the entire story, look into it deeper and things look worse. 66MHz x 32 bits, or 100MHz x 64 bits?

S.G., 19th October 2019, 18:10

Suppose there is a bandit who frequents the mountain pass between two countries. He robs the merchant caravans as they pass through, but is careful to take only as much as the merchants can afford to lose, so that they will keep using the pass and he will keep getting the loot.    A thief, right?    Now, suppose he has a license to charge tolls of those who use the pass, a license issued by the government of one of the countries -- or even both of them. The tolls are by coincidence equal to what he formerly took by force. How has the nature of his enterprise changed, simply through being made legal? He is still just a thief. He is still just demanding payment and not contributing anything in return. How can the mere existence of that piece of paper entitling him to rob the caravans alter the fact that what he is doing is in fact robbing them?    But now suppose instead of a license to steal, he has a land title to the pass. He now charges the caravans the exact same amount in "rent" for using the pass, and has become quite a respectable gentleman. But how has the nature of his business really changed? It's all legal now, but he is still just taking money from those who use what nature provided for free, and contributing nothing whatever in return, just as he did when he was a lowly bandit. How is he any different now that he is a landowner?    And come to that, how is any other landowner charging rent for what nature provided for free any different?

S.G., 19th October 2019, 18:13

Suppose I, knowing that a road would shortly be made into a particular region, bought from Government the privilege of placing bars across the road (when made) and forbidding anybody to pass until he had paid a toll; a toll not to pay for the maintenance of the road, but a toll for mere permission to pass along the road. Every one would recognise that this toll was pure blackmail and not earnings, and the obstructors mere parasites licensed to prey upon the public. But where is the difference between blocking the road and blocking the land that the roads lead to? Where is the difference between levying blackmail on the transport of goods and levying it on their production?    But I say, "It was with real earnings that I bought the right to demand this payment."    True. But the point is that whether I bought it or stole it, the thing I have bought or stolen is the privilege of levying blackmail upon industry; of demanding something and giving nothing in return; of laying my hand on the earth's surface and saying to all and sundry, "Give me of the produce of your labour, or be off with you; so much a year if I choose to let it; so much in a lump sum if I prefer to sell it." Whichever of the two forms the demand assumes it is called by political economists "rent," but it is not to be confounded with rent in the legal and commercial sense, which includes interest on the cost of improvements. The rent I shall mean is economic rent only; the price charged for the mere use of the land as such, either without any improvements or apart from them: "ground rent" in short.    Earnings mean taking payment for goods or service rendered; appropriation means taking something and giving nothing in return; no matter whether the taking be legal or illegal, or how I acquired the privilege of taking. Anyone can recognise that it is one thing to charge for the fish I caught in the sea, and quite another thing to charge for permission to fish in the sea; one thing to charge for produce I have raised from the land, and quite another thing to charge for permission to raise produce from land.

Rick, 19th October 2019, 18:48

S.G. - Stewart Goldwater?    If you have an axe to grind relating to people making money from stuff that you believe should be free, you're more than welcome to post it IN THE APPROPRIATE PLACE. Shoving it in the comments of a random article makes you look a bit dumb.  If you're leading up to suggesting Land Value Tax, we did that back in 2012...

Rick, 19th October 2019, 18:53

To save any future discussion on who owns what and who is thieving what - just refer to http://www.politicalforum.com/index.php?threads/the-problem-of-capitalism.552409/page-79#post-1071075237

S.G., 31st October 2019, 00:52

My posts were responses to your "We're not French or English, we're humans": I'm not grinding any axe!

Add a comment (v0.11) [help?] . . . try the comment feed!
		Your name
		Your email	(optional)
		Validation	Are you real? Please type 06397 backwards.
		Your comment