Intel Atom: pros/cons/hazzards?

On Tue, 23 Sep 2014 05:34:17 +0000 (UTC), glen herrmannsfeldt
<gah@ugcs.caltech.edu> wrote:

>In comp.arch.embedded rickman <gnuarm@gmail.com> wrote:
>
>(snip)
>
>> Your point about international sizes is likely valid.  I remember being 
>> surprised by the 750 ml size though.  Was that an international size 
>> long before the US converted to metric... in liquor sizes only of course.
> 
>> Are there any areas where the US uses metric by convention?
>
>As noted before, the duty free alcohol import limit is 1L.
>
>There might be many that have a scientific basis. Polution limits,
>and other chemical quantities might be metric. (Then again, they
>are othen ppm or ppb, I suppose those are metric fractions.)


Ahhh... pollution limits set by the EPA.  Check out page 23421
(actually the 9th page of the PDF) of:

http://www.gpo.gov/fdsys/pkg/FR-2014-04-28/pdf/2014-06954.pdf

You'll notice emission limits being given in milligrams per mile.

Crying at this point is perfectly understandable.

On Wed, 24 Sep 2014 08:15:54 -0500, Robert Wessel
<robertwessel2@yahoo.com> Gave us:

>On Tue, 23 Sep 2014 16:53:50 -0700, DecadentLinuxUserNumeroUno
><DLU1@DecadentLinuxUser.org> wrote:
>
>>On Tue, 23 Sep 2014 18:23:07 -0500, Robert Wessel
>><robertwessel2@yahoo.com> Gave us:
>>
>>>On Tue, 23 Sep 2014 09:51:12 -0700, DecadentLinuxUserNumeroUno
>>><DLU1@DecadentLinuxUser.org> wrote:
>>>
>>>>On Mon, 22 Sep 2014 23:27:22 -0400, rickman <gnuarm@gmail.com> Gave us:
>>>>
>>>>>Are there any areas where the US uses metric by convention?
>>>>
>>>>  All three car makers, and even AMC and Jeep, when they existed.
>>>>
>>>>  Since the early seventies.
>>>>
>>>>  The military, NASA, many other gov orgs, all metricized.
>>>
>>>
>>>NASA, only sort-of.  ISS is chock full of 5/16, 7/16 and 3/8 inch
>>>bolts.
>>
>>
>>  Nice snip of the subsequent statement which addressed the fact that
>>both are still used as the adoption of the new one does not cancel out
>>the use of the current one.  So, one will see mixed use.
>
>
>Little, if any, of the hardware on ISS was designed prior to the early
>seventies", when you said NASA went metric.  Another aerospace example
>is Boeing's rather new 787 - again, mostly standard sized parts.


  Again, you miss the point.  Both STILL get and WILL get used, even
within the same design.

  The adoption of metric does not cancel the continued use of the
previous standard.

  Several reasons cause this.  Just one that comes to mind would be the
implementation of COTS into the system(s).

  Look how long it has taken to stop using magnetic tape for data
storage.  It is, in fact, STILL being used, despite a changeout being
able to pay for itself in record time.

  So many elements of a modern society and governing body that have
grown so pathetically lethargic in refinement, all the while claiming to
be civil, and sophisticated and refined.

  I see no refinements.  I see undeserving greedy bastards everywhere.

In comp.arch.embedded DecadentLinuxUserNumeroUno <DLU1@decadentlinuxuser.org> wrote:

(snip)

>  Again, you miss the point.  Both STILL get and WILL get used, even
> within the same design.
 
>  The adoption of metric does not cancel the continued use of the
> previous standard.
 
(snip)

>  Look how long it has taken to stop using magnetic tape for data
> storage.  It is, in fact, STILL being used, despite a changeout being
> able to pay for itself in record time.

New magnetic tape systems are still being developed. 

To do so, and to make any sense, the storage density has to
keep increasing.  So far Ultrium (LTO) has been able to do that.

https://en.wikipedia.org/wiki/Linear_Tape-Open

LTO-6, released Dec 2012, stores 2.5TB uncompressed on one
cartridge. The data capacity doubles about every 2.5 years.

-- glen

On Wed, 24 Sep 2014 21:51:02 +0000 (UTC), glen herrmannsfeldt
<gah@ugcs.caltech.edu> Gave us:

>
>New magnetic tape systems are still being developed. 

  How utterly stupid of them.

Don Y <this@is.not.me.com> writes:

> Hi,
>
> I've rescued a couple of Atom-based SBC's.  Essentially
> diskless workstations.
>
> Am I correct in assuming these are just "yet-another-x86"?

Yes. I bought an Intel Atom motherboard, D510MO, to build a small file
server some years ago, for home use. The board including CPU sold for
just $60. Not really suited for a file server since it only has two SATA
ports, but back then PCI SATA cards were still easily available. OS is
bog standard Debian Linux for x86_64.

I run three disks, boot and two larger ones in RAID1 (Linux software
RAID). For slow file serving it's fine but there seem to be a rather
serious I/O bottleneck (disk to ram goes at 100MB/s, ram to net 80MB/s
but disk to net only 25 MB/s). I do sometimes run Firefox to get to my
router's config pages or some other light surfing but that's not much
fun. Firefox startup alone is about 10-15 seconds with CPU
pegged... Page rendering is painful.

I put it in a cheapie mini-ITX case. I had to add a case fan since the
cheapo case is so badly designed, the only front panel accessible place
for drives is right on top of the CPU heatsink... Quite a WTF
moment. Although with just one drive the case design is actually fine,
one disk goes on the bottom of the case and the case can act as a heat
sink for the drive.

> (Boxes in question are 1.6GHz, <4G DDR2, with all the
> typical "PC" I/O's)

Sounds a lot like the D510MO board. The CPU is an Atom D510, dual cores
at 1.6GHz, 4 GB DDR2 max, NM10 chipset. Legacy PC I/O, VGA, dual PS2
ports & serial. Second serial port and parallel port as pin headers on
the board.

In article <vg3y4t7hflk.fsf@coffee.modeemi.fi>, Anssi Saari  <as@sci.fi> wrote:

>I run three disks, boot and two larger ones in RAID1 (Linux software
>RAID). For slow file serving it's fine but there seem to be a rather
>serious I/O bottleneck (disk to ram goes at 100MB/s, ram to net 80MB/s
>but disk to net only 25 MB/s).

You may be saturating the PCI bus.

I've seen systems in which the primary PCI bus had only a limited
number of ports available, which were mostly used for "internal"
peripherals.  The PCI slots were all connected to secondary PCI bus,
which was connected to the primary via a PCI bridge.

I/O through the bridge was slow, with extra latency on each
transaction and a relatively small PCI transaction size.

As a result, when doing a multi-way copy (reading from one disk
connected to the primary-bus SATA controller, and writing "in
parallel" to four disks on a pair of two-port SATA PCI cards) I was
only getting about half of the throughput I expected.  An
older-generation motherboard, with a slower CPU, but with the PCI
slots on the primary PCI bus, was able to push data at about 95% of
the theoretical PCI-bus saturation speed.

You might want to check the bus configuration (use "lspci" if on
Linux), and try playing around with the PCI-bus latency and
transfer-size settings.

Also, for file-sharing of that sort, you might want to check your
filesystem readahead parameters.  Increasing the readahead size or
memory budget might improve matters.

On 9/25/2014 3:51 AM, Anssi Saari wrote:
> Don Y <this@is.not.me.com> writes:
>
>> I've rescued a couple of Atom-based SBC's.  Essentially
>> diskless workstations.
>>
>> Am I correct in assuming these are just "yet-another-x86"?
>
> Yes. I bought an Intel Atom motherboard, D510MO, to build a small file
> server some years ago, for home use. The board including CPU sold for
> just $60. Not really suited for a file server since it only has two SATA
> ports, but back then PCI SATA cards were still easily available. OS is
> bog standard Debian Linux for x86_64.

No PCI slots (of ANY flavor) in this box.  Truly an SBC.

> I run three disks, boot and two larger ones in RAID1 (Linux software
> RAID). For slow file serving it's fine but there seem to be a rather
> serious I/O bottleneck (disk to ram goes at 100MB/s, ram to net 80MB/s
> but disk to net only 25 MB/s). I do sometimes run Firefox to get to my
> router's config pages or some other light surfing but that's not much
> fun. Firefox startup alone is about 10-15 seconds with CPU
> pegged... Page rendering is painful.

Wow!  That's discouraging!  OTOH, I won't be using any display features
so any bottlenecks attributable to it would not apply.

> I put it in a cheapie mini-ITX case. I had to add a case fan since the
> cheapo case is so badly designed, the only front panel accessible place
> for drives is right on top of the CPU heatsink... Quite a WTF
> moment. Although with just one drive the case design is actually fine,
> one disk goes on the bottom of the case and the case can act as a heat
> sink for the drive.
>
>> (Boxes in question are 1.6GHz, <4G DDR2, with all the
>> typical "PC" I/O's)
>
> Sounds a lot like the D510MO board. The CPU is an Atom D510, dual cores
> at 1.6GHz, 4 GB DDR2 max, NM10 chipset. Legacy PC I/O, VGA, dual PS2
> ports & serial. Second serial port and parallel port as pin headers on
> the board.

Here's an image of the box:
<https://www.avforums.com/attachments/img_2334-jpg.444799/>

Unfortunately, the disk obscures a great deal.  :<

There's a second SATA port (power+signal) on the MB just under the blue
SATA data cable (upper left).  No idea how it is expected to be used as
to use it "as is" would result in the disk drive "aimed for the stars"
once mated to it!  (i.e., normal to the MB PCB)

Also obscured by disk are DVI & VGA connectors, some of the USB connectors
(6 on back, 2 on front) as well as single serial port (the PS2 mouse/keyboard
connectors are visible at the right of the disk).

Silver cooler front right appears to be for video (?).

AFAICT, no second serial port (at least not on any uncommitted headers).
Likewise, no parallel.  Any headers seem to be used to connect the
front panel to the MB (you can see a bundle of black-jacketed cables along
the left side of the image) or support the wireless module (under the disk).

It's a cute little box.  *With* a disk (and the fan hidden beneath it)
it is still virtually silent (i.e., I could easily sleep with this on
my night stand!).  Remove the disk (fan comes with it) and there's no
sound at all.

I may opt to configure one with a disk and PXE serve boot images to others.

Also pursuing Andrew's thumb drive approach.

Rescued an NAS the other day (primarily for the 2.5TB of media it contained)
so I can now clean off one of my (too large!) laptop SATA drives and get to
work exploring this.  Unfortunately, I can't imagine what to clutter up
a 640GB drive with in such a box!  :-/  (hence the idea of serving up images
to other boxen)

On 9/25/2014 1:07 PM, Don Y wrote:
> On 9/25/2014 3:51 AM, Anssi Saari wrote:
>
>> I run three disks, boot and two larger ones in RAID1 (Linux software
>> RAID). For slow file serving it's fine but there seem to be a rather
>> serious I/O bottleneck (disk to ram goes at 100MB/s, ram to net 80MB/s
>> but disk to net only 25 MB/s). I do sometimes run Firefox to get to my
>> router's config pages or some other light surfing but that's not much
>> fun. Firefox startup alone is about 10-15 seconds with CPU
>> pegged... Page rendering is painful.
>
> Wow!  That's discouraging!  OTOH, I won't be using any display features
> so any bottlenecks attributable to it would not apply.

If you use a browser, an older copy is better in that it uses less 
memory and runs faster.  Chrome is a lot better than Firefox in speed, 
but I understand it can use more memory. IE is the pits... lol

-- 

Rick

dplatt@coop.radagast.org (David Platt) writes:

> In article <vg3y4t7hflk.fsf@coffee.modeemi.fi>, Anssi Saari  <as@sci.fi> wrote:
>
>>I run three disks, boot and two larger ones in RAID1 (Linux software
>>RAID). For slow file serving it's fine but there seem to be a rather
>>serious I/O bottleneck (disk to ram goes at 100MB/s, ram to net 80MB/s
>>but disk to net only 25 MB/s).
>
> You may be saturating the PCI bus.
>
> I've seen systems in which the primary PCI bus had only a limited
> number of ports available, which were mostly used for "internal"
> peripherals.  The PCI slots were all connected to secondary PCI bus,
> which was connected to the primary via a PCI bridge.

Things are a little different since this is a PCIe chipset. There's a
decent block diagram in the motherboard manual at
http://downloadmirror.intel.com/18357/eng/D510MO_TechProdSpec02.pdf

So basically, SATA ports are connected directly to the chipset, the
integrated fast ethernet isn't used and instead an external Realtek
gigabit ethernet is connected via a PCIe x1 connection. Two PCI ports
exist and now that I think about it, transfers from SATA via PCI to
ethernet go faster than builtin SATA to ethernet...

So it looks like there's some internal bottleneck in the chipset from
SATA to PCIe but I don't really know what I could do. As is, 25 MB/s is
about 200 Mbits/s so way more than any video I might want to stream.
    
> You might want to check the bus configuration (use "lspci" if on
> Linux), and try playing around with the PCI-bus latency and
> transfer-size settings.

PCIe devices seem to always have 0 for latency. As for transfer size, I
can't seem to find anything about that.

> Also, for file-sharing of that sort, you might want to check your
> filesystem readahead parameters.  Increasing the readahead size or
> memory budget might improve matters.

Well, as above, for my file sharing this is plenty fast enough.

Hans-Bernhard Br�ker wrote:
> 
> Am 22.09.2014 um 19:06 schrieb Don Y:
> 
> > My point was that we all use an odd collection of units -- even "metrified"
> > countries -- every day.
> 
> You may never have been to a truly "metrified" country, then.
> 
> E.g. here in Germany there's basically no non-metric unit left in
> practical, much less every-day use.  And it's been like that since about
> the 1950s, at the least.


   None of your electronics is 19" rackmount?



-- 
Anyone wanting to run for any political office in the US should have to
have a DD214, and a honorable discharge.