[9fans] just an idea (Splashtop like)

[9fans] just an idea (Splashtop like)

[David Leimbach]

[-- Attachment #1: Type: text/plain, Size: 401 bytes --]

 A buddy of mine just got this: asus p5ql/epu motherboard.  It came with
Splashtop:  http://en.wikipedia.org/wiki/Splashtop
... which is a linux distribution that boots in like 5 seconds or so.
 Complete with BlackBox for a window manager, Skype, an instant messager
client and firefox.

I wonder if it could be changed to be a plan 9 terminal, or if one could at
least get 9vx on it.

Dave

[-- Attachment #2: Type: text/html, Size: 520 bytes --]

Re: [9fans] just an idea (Splashtop like)

[J.R. Mauro]

On Fri, Jul 31, 2009 at 1:50 PM, David Leimbach<leimy2k@gmail.com> wrote:
>  A buddy of mine just got this: asus p5ql/epu motherboard.  It came with
> Splashtop:  http://en.wikipedia.org/wiki/Splashtop
> ... which is a linux distribution that boots in like 5 seconds or so.
>  Complete with BlackBox for a window manager, Skype, an instant messager
> client and firefox.
> I wonder if it could be changed to be a plan 9 terminal, or if one could at
> least get 9vx on it.
> Dave

Doesn't ASUS burn the Linux distro into a chip, though? Maybe there
are utilities to flash it with something else.

Re: [9fans] just an idea (Splashtop like)

[ron minnich]

On Fri, Jul 31, 2009 at 6:56 PM, J.R. Mauro<jrm8005@gmail.com> wrote:

> Doesn't ASUS burn the Linux distro into a chip, though? Maybe there
> are utilities to flash it with something else.

see flashrom at coreboot.org

This is a great idea assuming we can get a mobo that plan 9 can use.

ron

Re: [9fans] just an idea (Splashtop like)

[erik quanstrom]

On Fri Jul 31 23:13:51 EDT 2009, rminnich@gmail.com wrote:
> On Fri, Jul 31, 2009 at 6:56 PM, J.R. Mauro<jrm8005@gmail.com> wrote:
>
> > Doesn't ASUS burn the Linux distro into a chip, though? Maybe there
> > are utilities to flash it with something else.
>
> see flashrom at coreboot.org
>
> This is a great idea assuming we can get a mobo that plan 9 can use.

why not just use normal hardware and buy a DOM for it?
none of coraid's machines, including the fileserver (!) have
spinning media.  only the srs (coraid storage appliances) do.

you can get a 128mb pata dom for $13 and sata for $30
at memorydepot.com (not an endorsement but i've purchased
from them before.)

i would imagine that regular mb + dom would be competitive
with the asus boards.

- erik

Re: [9fans] just an idea (Splashtop like)

[tlaronde]

On Sat, Aug 01, 2009 at 07:58:10AM -0400, erik quanstrom wrote:
>
> why not just use normal hardware and buy a DOM for it?
> none of coraid's machines, including the fileserver (!) have
> spinning media.  only the srs (coraid storage appliances) do.
>
> you can get a 128mb pata dom for $13 and sata for $30
> at memorydepot.com (not an endorsement but i've purchased
> from them before.)

Well, if the DOM is compatible with traditionnal IDE it could be, too,
an option for installation: an easily pluggable HD instead of floppy or
CD.

I didn't even know that this exists... so thanks for the tip, Erik!
--
Thierry Laronde (Alceste) <tlaronde +AT+ polynum +dot+ com>
                 http://www.kergis.com/
Key fingerprint = 0FF7 E906 FBAF FE95 FD89  250D 52B1 AE95 6006 F40C

Re: [9fans] just an idea (Splashtop like)

[erik quanstrom]

> Well, if the DOM is compatible with traditionnal IDE it could be, too,
> an option for installation: an easily pluggable HD instead of floppy or
> CD.

the dom is compatable with traditional ide, and with the distributed
sdata.c.  before i found the doms, i used a pata <-> cf bridge.

- erik

Re: [9fans] just an idea (Splashtop like)

[ron minnich]

btw the sata FLASH parts are surprisingly fast. Not at all like USB
sticks, if that is what you are used to.
ron

Re: [9fans] just an idea (Splashtop like)

[J.R. Mauro]

On Sat, Aug 1, 2009 at 11:49 AM, ron minnich<rminnich@gmail.com> wrote:
> btw the sata FLASH parts are surprisingly fast. Not at all like USB
> sticks, if that is what you are used to.
> ron
>
>

Ron, have you researched any long-term wear studies on these flash
drives? I've heard a lot of good things,
but I'm really put off by terms like "wear levelling", filesystems
optimized to work around flash's delicateness,
etc.

I'm really interested in any numbers anyone has.

Re: [9fans] just an idea (Splashtop like)

[John DeGood]

J.R. Mauro wrote:
> Doesn't ASUS burn the Linux distro into a chip, though? Maybe there
> are utilities to flash it with something else.

I believe new ASUS motherboards typically boot Splashtop from hard disk,
not from flash.  On my ASUS M4A78 PLUS, Splashtop can be installed into
a directory tree "\ASUS.SYS" (~500 MB) on any partition with a Windows
filesystem.  The ASUS BIOS displays a gaudy splashscreen that allows the
user to choose Splashtop or the normal OS.

Source seems to be here:

  http://www.splashtop.com/open_source.php

John

Re: [9fans] just an idea (Splashtop like)

[J.R. Mauro]

On Sat, Aug 1, 2009 at 10:55 PM, John DeGood<bulk@degood.org> wrote:
> J.R. Mauro wrote:
>> Doesn't ASUS burn the Linux distro into a chip, though? Maybe there
>> are utilities to flash it with something else.
>
> I believe new ASUS motherboards typically boot Splashtop from hard disk,
> not from flash.

Thanks for setting that straight -- my reference was a couple of old
articles with wording that Asus was going to put Splashtop on their
computers in flash.

Obviously the articles I remember are long obsolete.

Re: [9fans] just an idea (Splashtop like)

[erik quanstrom]

> Ron, have you researched any long-term wear studies on these flash
> drives? I've heard a lot of good things,
> but I'm really put off by terms like "wear levelling", filesystems
> optimized to work around flash's delicateness,
> etc.
>
> I'm really interested in any numbers anyone has.

just looking at the intel x25-e datasheet, the URE rate
(unrecoverable read error) is the same as enterprise sata
drives at 1e-15, but the mtbf is higher, but within a factor
of two.

assuming honest mtbf numbers, one would expect similar
ures for the same io workload on the same size data set
as mechanical disks.  since flash drives are much smaller,
there would obviously be fewer ures per drive.  but needing
10x more drives, the mtbf would be worse per byte of storage
than enterprise sata drives.  so you'd see more overall failures.

conclusion: you'll need raid for flash drives, too.

this is a pretty suprising result.  and i'm sure that a large
number of people are going to jump up and argue.  but
here are the datasheets.

http://download.intel.com/design/flash/nand/extreme/319984.pdf
http://www.wdc.com/en/library/spec/2879-701281.pdf
(i didn't see the wdc mtbf but i've seen it quoted as 1.2Mhrs, as
http://hothardware.com/News/WD-Introduces-RE3-Enterprise-SATA-Hard-Drives/
)

perhaps the reason that it's so suprising is the same reason
we didn't pay attention to the ure rate when hard drives were
512mb.  would you expect to have a bad spot in 2,000 fujitsu
eagles?  that's ~ the amount of data you can store on one tb drive.

- erik

Re: [9fans] just an idea (Splashtop like)

[Anthony Sorace]

On Sun, Aug 2, 2009 at 14:17, erik quanstrom<quanstro@coraid.com> wrote:

> assuming honest mtbf numbers, one would expect similar
> ures for the same io workload on the same size data set
> as mechanical disks.  since flash drives are much smaller,
> there would obviously be fewer ures per drive.  but needing
> 10x more drives, the mtbf would be worse per byte of storage
> than enterprise sata drives.  so you'd see more overall failures.

this depends on usage, obviously. i think it misses the point that
there's plenty of applications where the smaller storage (assuming a
single unit) is perfectly adequate. i swapped out the HD in my laptop
for a SD drive: the reduction in size is entirely workable, and the
other benefits make the trade a big win. there're plenty of
applications where i need relatively little raw storage: laptops, boot
media for network terminals, embedded things.

for large-scale storage, your analysis is much more appropriate. my
file server remains based on spinning magnetic disks, and i expect
that's likely to be the case for a long time.

Re: [9fans] just an idea (Splashtop like)

[erik quanstrom]

> > assuming honest mtbf numbers, one would expect similar
> > ures for the same io workload on the same size data set
> > as mechanical disks.  since flash drives are much smaller,
> > there would obviously be fewer ures per drive.  but needing
> > 10x more drives, the mtbf would be worse per byte of storage
> > than enterprise sata drives.  so you'd see more overall failures.
>
> this depends on usage, obviously. i think it misses the point that
> there's plenty of applications where the smaller storage (assuming a
> single unit) is perfectly adequate. i swapped out the HD in my laptop
> for a SD drive: the reduction in size is entirely workable, and the
> other benefits make the trade a big win. there're plenty of
> applications where i need relatively little raw storage: laptops, boot
> media for network terminals, embedded things.
>
> for large-scale storage, your analysis is much more appropriate. my
> file server remains based on spinning magnetic disks, and i expect
> that's likely to be the case for a long time.

on the other hand, since the ure rate is the same for a
mechanical disk as for your flash drive, one can't claim that
it's "more reliable".  it will return an unreadable error just
as often.  limiting your dataset on a mechanical hard drive
would accomplish the same goal for less cash.  and the afr (dirty secret:
the mtbf number is actually the extrapolated afr^-1) is only
0.4% instead of 0.7%.  at that rate, something else is more
likey to eat your laptop (gartner sez 20%/year.  but that's for
the intel enterprise ssd, which costs more than most laptops.
this article claims that flash is currently less reliable
the old-fashoned disks:

http://www.pcworld.com/article/143558/laptop_flash_drives_hit_by_high_failure_rates.html

surprising, no?  there are still plenty of reasons to want an
ssd.  it just seems that reliablity isn't one of those reasons yet.

- erik

Re: [9fans] just an idea (Splashtop like)

[Devon H. O'Dell]

2009/8/2 erik quanstrom <quanstro@quanstro.net>:
> http://www.pcworld.com/article/143558/laptop_flash_drives_hit_by_high_failure_rates.html
>
> surprising, no?  there are still plenty of reasons to want an
> ssd.  it just seems that reliablity isn't one of those reasons yet.

The big one for me has always been that I tend to drop the sort of
machines while they're running. Does a lot to the MTBF for standard
drives... not so much for SSD.

--dho

> - erik
>
>

Re: [9fans] just an idea (Splashtop like)

[erik quanstrom]

> 2009/8/2 erik quanstrom <quanstro@quanstro.net>:
> > http://www.pcworld.com/article/143558/laptop_flash_drives_hit_by_high_failure_rates.html
> >
> > surprising, no?  there are still plenty of reasons to want an
> > ssd.  it just seems that reliablity isn't one of those reasons yet.
>
> The big one for me has always been that I tend to drop the sort of
> machines while they're running. Does a lot to the MTBF for standard
> drives... not so much for SSD.

doesn't this make the pcworld figures even more surprising?

- erik

Re: [9fans] just an idea (Splashtop like)

[John DeGood]

erik quanstrom wrote:
> just looking at the intel x25-e datasheet, the URE rate
> (unrecoverable read error) is the same as enterprise sata
> drives at 1e-15, but the mtbf is higher, but within a factor
> of two.
>
> assuming honest mtbf numbers, one would expect similar
> ures for the same io workload on the same size data set
> as mechanical disks.
...
> here are the datasheets.
>
> http://download.intel.com/design/flash/nand/extreme/319984.pdf
> http://www.wdc.com/en/library/spec/2879-701281.pdf

For the Intel SSD one must also consider:

> 3.5.4 Write Endurance
> 32 GB drive supports 1 petabyte of lifetime random writes and 64 GB drive supports 2 petabyte of lifetime random writes.

That is equivalent to writing the capacity of the SSD 31250 times.  At
the specified random 4K write rate of 3300 IOPS one could wear out the
SSD in 876 days.  Non-random writes could cause more rapid wear,
depending on their pattern and the wear leveling algorithms in the SSD.

John

Re: [9fans] just an idea (Splashtop like)

[erik quanstrom]

> For the Intel SSD one must also consider:
>
> > 3.5.4 Write Endurance
> > 32 GB drive supports 1 petabyte of lifetime random writes and 64 GB drive supports 2 petabyte of lifetime random writes.
> That is equivalent to writing the capacity of the SSD 31250 times.  At
> the specified random 4K write rate of 3300 IOPS one could wear out the
> SSD in 876 days.  Non-random writes could cause more rapid wear,
> depending on their pattern and the wear leveling algorithms in the SSD.

do you think this is a serious limitation?  by my calculation, assuming
that you read everything written at least once and 10x faster read than
write leading to 3300 iops taking 1.1s
	1000^5 bytes /(3300 s^-1 * 1.1^-1 * 4*1024 bytes)/86400s/day
		= 942 days
this is 153 days short of the product lifetime.  by the way, one
would expect ~8 ures during this test (8e15 bits/1 ure/1e-15 bits).
(http://download.intel.com/support/ssdc/hpssd/sb/english_ssd_3_year_warranty.pdf)

do you really think its reasonable that someone could run this
drive at 100% of capacity for 2½ years?  even allowing for shipping
and installation time will get you pretty close to the warranty.
can you think of how this could be done with a plan 9 application
that's doing something useful?

it's hard to know if non-random writes create more wear than intel
specifies or not.  strictly sequential i/o should create similar wear
because 16 4k writes can be combined into one flash cycle and
16*3300*4k is about 216 mb/s.  so i don't see how you can get in
more flash cycles than 3300/s and increase the wear rate.

- erik

Re: [9fans] just an idea (Splashtop like)

[John DeGood]

erik quanstrom wrote:
> do you really think its reasonable that someone could run this
> drive at 100% of capacity for 2½ years?  even allowing for shipping
> and installation time will get you pretty close to the warranty.
> can you think of how this could be done with a plan 9 application
> that's doing something useful?
>
> it's hard to know if non-random writes create more wear than intel
> specifies or not.  strictly sequential i/o should create similar wear
> because 16 4k writes can be combined into one flash cycle and
> 16*3300*4k is about 216 mb/s.  so i don't see how you can get in
> more flash cycles than 3300/s and increase the wear rate.

Short flash lifetimes in some consumer products have been attributed to
poor or even nonexistent wear leveling algorithms.  If the wear leveling
algorithms used in SSDs are very good and the usage pattern doesn't
tickle some pathological case, then flash wear will probably not be an
issue.  I look forward to someone like Google or one of the DoE sites
deploying large numbers of SSDs and publishing reliability statistics.

John

Re: [9fans] just an idea (Splashtop like)

[Dave Eckhardt]

> would you expect to have a bad spot in 2,000 fujitsu eagles?

If you do, I have a repair manual.

Dave Eckhardt

Re: [9fans] just an idea (Splashtop like)

[erik quanstrom]

On Sat Aug  8 00:47:40 EDT 2009, davide+p9@cs.cmu.edu wrote:
> > would you expect to have a bad spot in 2,000 fujitsu eagles?
>
> If you do, I have a repair manual.

i prefer two strong oxen.

- erik