Re: [NTLK] ATA Linear flash speed

On Aug 19, 2008, at 10:37 AM, Ryan Vetter wrote:

> For me, since I don't really, use
> my Newton over the Internet
> downloading stuff from the web or
> writing to another linear card
> is not something I would be using the
> Newton for.Ê

Hmmm...

When you finally have an ATA card that works, you tend to want to put some
stuff on it (e.g. large eBooks, a movie or two, multiple apps, editor
enhancements... like dictionaries, etc.) Yes? No?

If you narrow it down, there are only 4 ways anyone would do this:
 1) Serial or LocalTalk
 2) Wired or Wireless
 3) Copy from existing stores
 4) Beam it

So... which of these speeds would you like to know about? If you get bored
easily, stop reading RIGHT NOW and skip to the last two paragraphs. I type
50 - 70 WPM and I feel like typing.

So... Are you ready contestants? Let's play find the bottleneck!

1) Serial or LocalTalk

The standard serial port speed (when connecting to NCU) is 38,400bps,
whether you are using a Mac or Windows. This is a fact. Especially if you
are classified as a noob. Got a KeySpan serial port adapter? Sorry bud,
you're also at 38,400bps when connecting to NCU. However, if you are
experienced, we can avoid any arguments and agree that you might be able to
hack your serial speed to 57,600bps.

Let's ask our studio audience, "Anybody brave enough to get 115,200bps? How
about 230,400bps?" If you search the archives, you will find plenty of
posts asking what to do, to prevent being disconnected when trying large
transfers, with any 'serial' speed beyond the default.

With AppleTalk over serial (a.k.a. LocalTalk), a speed of 230,400bps is
actually attainable! LocalTalk uses a special chip and an HDLC packet based
protocol with hardware error checking. This 'true' LocalTalk type of serial
connection can, and does, operate at near theoretical max. It is only
available using a Classic Mac with (an on board) 8-pin DIN 'Printer' port.
You must have 'AppleTalk' active on that 'Printer' port and then you can
get a true 230,400bps native connection to NCU... without hacks.

What exactly do these numbers mean? It means the maximum throughput you can
expect from the 'average' serial connection is (roughly) 3-5K Bytes per
second. Can the Newton keep up with that? Absolutely and without much
effort. But a 12MB backup file will take close to an hour to complete.

The Newton was actually designed to work really well with LocalTalk over
serial. If you haven't investigated it, it's pretty darned fast... about
28K Bytes/second. I would suspect that 28K-30K is about the fastest that
the Newton will handle any data. In, out, on, off, here, there, or
anywhere.

2) Wired or Wireless

Here is were we start hitting some ceilings. Forget about 10Mbps. Let's
make things simple and talk about 1Mbps. Let's see, 1 million, divided by 8
(bits to a byte) and then by 1024... equals a maximum estimate of
122KB/sec. Almost DSL speed.

Is 122K really possible on a Newton? Nope. The best I've seen (wirelessly)
was about 30KB/sec during a backup. Keep in mind that a wireless backup is
almost an optimal situation: Very high bandwidth available to a (primarily)
solid read operation.

What about write operations? Well, things start to really slow down when
the the Newton needs to write. Keep in mind, were talking Internal Flash
and Linear Flash cards only. We're not going to add ATA to the equation
quite yet. The maximum speed I could ascertain for a restore via wireless
was about 11KB/sec. That means that you can estimate that between 10KB/sec
and 30KB/sec is about the best that a Newton can communicate with another
Host (NCU/NCX), and still keep up with all the data coming down the pipe.

3) Copy from existing stores

Alright, now let's see what we can do with straight hardware. When using
the internal flash RAM with an external linear flash card, how fast can the
data go from internal to linear and linear to internal operations? Well, I
believe I've answered that one already.

On Aug 19, 2008, at 5:01 AM, R A Parker wrote:

> Internal to Linear
> 409K 17 Seconds 24.059KB/sec
> Linear to Internal
> 409K 23 Seconds 17.782KB/sec
>

Hmmm... right on target. The Newton can do about 17KB/s - 24KB/s, when
moving data around on it's internal hardware.

4) Beaming it
Forget it... I've only seen about 3KB/s and typically, beaming is very
unstable with larger packages.

Now, lets add an ATA card to the mix. Take a look at the speeds I posted
for ATA data transfers.

> First test: Downloading this movie
> from NCU in Classic Mode via
> AppleTalk Wifi Bridge on my G4/500AGP.
>
> Wifi to Linear Flash:
> 2,389K 182 Seconds 13.126KB/sec
> Wifi to ATA Card:
> 2,389K 2,940 Seconds 0.891KB/sec
>
> LINEAR/ATA DIFFERENCE:
> 6.19% 14.73 times faster
>
>
> Second Test: Filing the same movie
> from one store to another.
>
> Filing ATA to Linear:
> 2,232K 73 Seconds 30.575KB/sec
> Filing Linear to ATA:
> 2,232K 603 Seconds 3.701KB/sec
>
> LINEAR/ATA DIFFERENCE:
> 12.11% 8.261 times faster
>
>
> Third Test: Filing the 1984.pkg from
> one ATA partition to another ATA
> partition, within the same card.
>
> Filing from ATA to ATA:
> 2,232K 390 Seconds 5.723KB/sec
>
>

You should be able to get a good idea, just what a Newton is capable of?

1) Internal Flash RAM appears slower than External Flash
2) Write operations are slower than read operations
3) Maximum data handling for a Newton with a Linear Flash is about 10KB/sec
to 30KB/sec.

> What interests me is the everyday speed.Ê

Let's see what Paul says,

On Aug 19, 2008, at 12:20 AM, Paul Guyot wrote:

> ATA and linear cards work very
> differently. Globally, ATA cards
> feel slower than linear cards:
> ATA cards can be faster in some
> (very) specific scenarios, but not
> the ones when you traditionally wait
> (i.e. package installation).
>
> There is something that ATA cards do
> very bad,

<snip>

> It's the handling of LBO, large
> binary objects, especially within
> transactions. This is typically what
> happens when you install packages.
>

So, as Paul eluded to, and I've experience, everyday speeds are not that
bad. HWR and other Newton OS operations are actually happening live,
character by character without delay. Moving small chunks of data around,
like notes, dates or other records, will, on rare occasions, feel about 50%
of normal because there are occasional pauses (a perceptible delay) when
the ATA partition is accessed.

Everyday type of operations and some data is so small you really can't tell
the difference between what is "normal" and what is "slowed down" by ATA.
Typically, there is not much big stuff (i.e. LBO's) being transferred to
and from the stores, during normal use. Of course, that would be dependent
on which application you are using.

So far, I've only tested moving packages (especially big ones) on to, from,
and around my ATA drive. Next, I'm going try out Internet browsing with the
ATA set as my cache drive. After that, email downloading with an ATA
partition set as the default storage location. After that, I want to test
GPS with live, auto tracking, graphic background maps. All of these should
be very data intense and will probably be impacted greatly, in some way, by
the slow nature of ATA. I'll let ya know.

> I did write some packages to the ATA
> and it did not seem that slow.

Yes, I would say that smaller packages may appear to be just as fast. When
we're talking about the difference of micro-seconds, you might not be aware
of ATA's impact.

> I think that would be a good test:
> writing package to the ATA
> and comparing its speed to
> the Linear card.

Did that. Typical (LBO) data operations (on ATA) appear to be about 6% -
13% of the speed when compared to Linear flash.

In fact, I've just finished loading my new 150x CF ATA drive's primary 4MB
partition. It is set as an Automount "Editor" partition. It contains
HyperNewt, Merlin/WordSleuth Spell Check & Thesaurus, MacinTalk Speech, and
other types of editing must haves. Some of these things are big, some
small.

Also of note, I've type and edited this entire email on my eMate which had
(by default) stored the doc on the ATA partition. I then spell checked the
whole thing and had it spoken back to me as well. Although the doc was
stored to the ATA, I was still able to type 50-70wpm without a problem. The
spell check didn't seem all that slow either. And, MacinTalk Text-to-Speech
only had one brief electronic skip while the Newton was reading it to me.

I would say that normal (everyday) usage of a Newton, with a High Speed ATA
CF card and adapter, is hardly perceptible as being slower than what I'm
used to with Linear Flash. The greatest thing I've noticed is how much more
stable the 150x (High Speed) CF Card is compared to the spare older,
slower, 256MB, SanDisk CF & SD Cards I was trying to use. An amazing
difference!

There... was that the answer you were waiting for? Sorry it took so long to
get there.

Sincerely,
Ron A. Parker

--
Sent using GoFetch and a Wireless Newton 2100.
--
RAParker
QuadzillaNET
       |\/|\ @ Newted.ORG
       |/-|/ @ Quadzilla.NET
       |\ |  @ TahoeSunsets.COM
--
Fetch...Send...Stay... "Good Boy!"
--
==================================================================== 
The NewtonTalk Mailing List - http://www.newtontalk.net/
The Official Newton FAQ     - http://www.splorp.com/newton/faq/
The Newton Glossary         - http://www.splorp.com/newton/glossary/
WikiWikiNewt                - http://tools.unna.org/wikiwikinewt/
====================================================================