[9fans] cheep ssds

[9fans] cheep ssds

[erik quanstrom]

i just got a 40gb intel ssd for building up a new file
server.  i really wasn't expecting 275 out of a theoretical
300mb/s.  maybe i forgot how to count and it's actually
27.5.  :-)

blakely# time dd -if /dev/sdE0/data -of /dev/null -bs 512k -count 100
100+0 records in
100+0 records out
0.00u 0.01s 0.19r 	 dd -if /dev/sdE0/data -of /dev/null ...
blakely# hoc
512*1024*100/0.19
275941052.6316

- erik

Re: [9fans] cheep ssds

[dexen deVries]

On Wednesday 03 of August 2011 05:53:51 erik quanstrom wrote:
> i just got a 40gb intel ssd for building up a new file
> server.  i really wasn't expecting 275 out of a theoretical
> 300mb/s.  maybe i forgot how to count and it's actually
> 27.5.  :-)

ah, marketing. megabits per second, rather than megabytes.

nb., SATAII isn't twenty times faster than PATA, just two times. 3GBit minus 
8/10 encoding.

-- 
dexen deVries

[[[↓][→]]]

For example, if the first thing in the file is:
   <?kzy irefvba="1.0" rapbqvat="ebg13"?>
an XML parser will recognize that the document is stored in the traditional 
ROT13 encoding.

(( Joe English, http://www.flightlab.com/~joe/sgml/faq-not.txt ))

Re: [9fans] cheep ssds

[erik quanstrom]

> On Wednesday 03 of August 2011 05:53:51 erik quanstrom wrote:
> > i just got a 40gb intel ssd for building up a new file
> > server.  i really wasn't expecting 275 out of a theoretical
> > 300mb/s.  maybe i forgot how to count and it's actually
> > 27.5.  :-)
>
> ah, marketing. megabits per second, rather than megabytes.
>
> nb., SATAII isn't twenty times faster than PATA, just two times. 3GBit minus
> 8/10 encoding.

i would think it takes some concerted mental effort to rain on a 275mb/s parade.
91% efficiency is near staturation.  :-).  i remember when sata 3.0gbps was overkill.
that was 2009.

i think the speed ratio of 2:1 is not correct.

first, you must be assuming udma 133.  i have never seen a udma 133 target.
	http://www.newegg.com/Product/ProductList.aspx?Submit=Property&Subcategory=14&Description=&Type=&N=100007603&IsNodeId=1&srchInDesc=&MinPrice=&MaxPrice=&PropertyCodeValue=359%3A7788

the fastest udma device i can find is udma 100.  this is not 100mhz, it's
100mb/s.  where as 3.0gb sata is 3.0gb/s / 10 bits/byte = 300mb/s.
that's 3x, not 2x.  (both are in decimal, as you can calculate from the udma
cycle time of 40ns, double-pumped).  and of course sata 6.0gb is
6x faster than pata.

that being said, pata is dead as fried chicken.  it's worse in every way than sata.
(no hot plug, shared channels, terrible connector, etc, that master vs slave
nonsense.)  i could go on.  i'd use sata over pata even if it cost me a bit
of speed.

- erik

Re: [9fans] cheep ssds

[dexen deVries]

On Wednesday 03 of August 2011 15:16:00 erik quanstrom wrote:
> i think the speed ratio of 2:1 is not correct.
> 
> first, you must be assuming udma 133.  i have never seen a udma 133 target.
> 	
http://www.newegg.com/Product/ProductList.aspx?Submit=Property&Subcategory
> =14&Description=&Type=&N=100007603&IsNodeId=1&srchInDesc=&MinPrice=&MaxPric
> e=&PropertyCodeValue=359%3A7788
> 
> the fastest udma device i can find is udma 100.  this is not 100mhz, it's
> 100mb/s.  where as 3.0gb sata is 3.0gb/s / 10 bits/byte = 300mb/s.
> that's 3x, not 2x.  (both are in decimal, as you can calculate from the
> udma cycle time of 40ns, double-pumped).  and of course sata 6.0gb is
> 6x faster than pata.

thanks for the correction and please excuse my post; it was an early-morning 
brain-fart :^)

-- 
dexen deVries

[[[↓][→]]]

For example, if the first thing in the file is:
   <?kzy irefvba="1.0" rapbqvat="ebg13"?>
an XML parser will recognize that the document is stored in the traditional 
ROT13 encoding.

(( Joe English, http://www.flightlab.com/~joe/sgml/faq-not.txt ))

Re: [9fans] cheep ssds

[Jack Norton]

erik quanstrom wrote:
> i just got a 40gb intel ssd for building up a new file
> server.  i really wasn't expecting 275 out of a theoretical
> 300mb/s.  maybe i forgot how to count and it's actually
> 27.5.  :-)
>
> blakely# time dd -if /dev/sdE0/data -of /dev/null -bs 512k -count 100
> 100+0 records in
> 100+0 records out
> 0.00u 0.01s 0.19r 	 dd -if /dev/sdE0/data -of /dev/null ...
> blakely# hoc
> 512*1024*100/0.19
> 275941052.6316
>
> - erik
>

I'd be curious on an update in the future concerning reliability.  I
think you of all people could give it a good workout.  I'd love to put a
cache (or a worm when prices come down on high capacity guys) on an ssd
in plan 9, but I am expecting it to quit after a few months for some
reason...  I'm not very trusting for some reason.

-Jack

Re: [9fans] cheep ssds

[erik quanstrom]

> I'd be curious on an update in the future concerning reliability.  I
> think you of all people could give it a good workout.  I'd love to put a
> cache (or a worm when prices come down on high capacity guys) on an ssd
> in plan 9, but I am expecting it to quit after a few months for some
> reason...  I'm not very trusting for some reason.

most of my servers right now (~20 of them) are using a small ssd
for nvram and a few other things.  in that capacity, i haven't had
any failures.  nor have i seen any failures personally using them as primary
storage.  i have killed a two by messing around, but that was clearly
operator error.  (unfortunately, there were witnesses.)

if you're using a layout similar to ken's fs, the cache access pattern
should be nearly sequential.  it's the worm were ssds will come into
their own, since the access pattern is quite random.

my current worm was initialized in 2006 and i'm using just 14.4gb.
it would probablly be cost effective to switch to ssds at this point.

- erik

Re: [9fans] cheep ssds

[Charles Forsyth]

i've had one potentially important ssd fail within 15 hours of my copying
stuff onto it; it failed completely (ie, the bios hung up probing it).
the drive was marked "refurbished", which makes me think i was rather a victim
of an experiment (``can we refurbish ssds that have previously shown a fatal failure?'' ... `perhaps not!').
otherwise, i've had nothing but luck with ssds: they are fast, quiet, and withstand movement.