SCIENTIFIC-LINUX-USERS Archives

July 2014

SCIENTIFIC-LINUX-USERS@LISTSERV.FNAL.GOV

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Konstantin Olchanski <[log in to unmask]>
Reply To:
Konstantin Olchanski <[log in to unmask]>
Date:
Fri, 11 Jul 2014 19:25:54 -0700
Content-Type:
text/plain
Parts/Attachments:
text/plain (140 lines)
On Fri, Jul 11, 2014 at 06:52:07PM -0700, ToddAndMargo wrote:
> On 07/11/2014 06:12 PM, Konstantin Olchanski wrote:
> >On Fri, Jul 11, 2014 at 04:02:09PM -0400, Andrew Z wrote:
> >>>
> >>>... synchronizing a flashing drive (target) with my hard drive (source) ...
> >>>
> >>>Problem: it is slow -- takes three hours.  To help the
> >>>speed issue, I upgraded from USB 2 to USB 3.  Backup went
> >>>from 3 hr-15 min to 3 hr-5 min.  It is almost faster
> >>>to wipe the stick and rewrite it.
> >>>
> >
> >
> >The main question is this: what is the actual write speed of your USB flash media? How about the re-write speed?
>  (not the same, obviously - as it requires as erase step).
> 
> Kanguru SS3:  Read 105 MB/sec;  78 MB/sec
> 


Right, this is almost SSD quality flash, but for some reason, performance for writing small files
is much worse compared to SSDs. One would think that one could make a USB flash disk
with same performance as SATA flash (SSD), but perhaps nobody makes the right
flash controller chip and a dual chip solution is not workable (an SSD flash controller
behind a USB-to-SATA interface).

More likely, though, is that the USB form factor, power budget and cooling capacity
does not permit implentation of full-performance flash disks.

One would think that USB3 can provide enough juice, but maybe there are still
problems with cooling and maybe they do not want to make a device that would
not work in a USB2 slot...


K.O.



>
> Don't have the re-write speed.
> 
> >
> >I ask because I use USB flash media as boot and linux system disks on embedded machines (VME SBCs)
> >and I have looked at different USB flash media. Most of them are very slow, actually it is very hard
> >to find "fast" USB flash media.
> 
> The cheap ones crawl.  If you spend a little:
> 
>    El-Cheap-O:
>        Read:  3 MB/sec
>        Write: 2 MB/sec
> 
>    Kanguru SS3:
>        Read:  105 MB/sec
>        Write: 78 MB/sec
> 
>    Kanguru Flash Blu 30:
>        Read:  145 MB/sec
>        Write: 8GB: 25 MB/sec
>               16,32,64GB: 45 MB/sec
> 
>    Kingston Data Travler:
>        Read:  70 MB/sec
>        Write: 30 MB/sec
> 
>    Kingston Data Travler Hyper-X 3.0:
>        Read:  225 MB/sec
>        Write: 135 MB/sec
> 
> For comparison:  IntelSSD 530 Series  SATA 3 flash drives:
>     Sequential Read:   540 MB/s
>     Sequential Write:  490 MB/s
> 
> >The common media you get for $10 at Staples is "read 30M/s, write 10M/s" (regardless of USB2 or USB3 interface).
> 
> Worse than that!
> 
> > This is probably consistent with the speeds that you see.
> 
> 
> I went from 20 MB/sec to 78 MB/sec.  Took 10 minutes off of
> three hours.
> 
> >
> >With some work, you can find media that writes at 20-30M/s, as measured by timing "dd", but drops
> >severely when you time "rsync" (must be inefficient at writing small files).
> >
> >So when you select a brand USB flash drive for your workload, as you run "rsync", watch
> >the output of "vmstat 1" (the "bo" column is Mbytes/sec written to disk) and
> >the output of "iostat -x 1" - you will see %util pegged at 100% and "svctm" (in msec) running
> >in 1-10-20 seconds for "slow media", a little bit smaller for betyter media. For HDDs and SSDs,
> >the "svctm" is in low milliseconds. "svctm" is the request service time - time from sending
> >a request to the drive and getting the reply from the drive that the request is finished.
> >
> >Some USB drives advertize high write speeds (not "up to" but actual "will write at" promises),
> >you can try those ($$$), but you will probably find that the speed of "rsync" does not reach
> >the promised rates because of inefficiency of flashing small files.
> 
> sda is the source hard drive; sdc is the target flash drive
> 
> $ iostat -x 1
> 
> Device:         rrqm/s   wrqm/s     r/s     w/s   rsec/s   wsec/s
> avgrq-sz avgqu-sz   await  svctm  %util
> 
> sda               0.00     0.00   28.00    0.00  7168.00     0.00
> 256.00     0.08    2.79   1.39   3.90
> 
> sdc               0.00     0.00    0.00  280.00     0.00  7406.00
> 26.45     1.34    4.74   3.25  90.90
> 
> 
> Couldn't find "svctm".
> 
> Does the above tell you anything?
> 
> 
> >
> >P.S. Another problem with USB flash drives - all brands except for 1 or 2 do not survive
> >being used as linux system disks - they brick themselves within days or weeks. I notice
> >that they tend to run quite hot, so I suspect they simply overheat and die. Actually,
> >I did not find a single USB3 flash drive that survives use as linux system disk yet.
> >By luck I have enough Patriot RageXT 8GB and 16GB USB2 flash media, these seem to last.
> 
> There is a lot of trash out there.  I refuse to sell Buffalo,
> as  have had almost 100% returns on them.
> 
> 
> -- 
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> Computers are like air conditioners.
> They malfunction when you open windows
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

-- 
Konstantin Olchanski
Data Acquisition Systems: The Bytes Must Flow!
Email: olchansk-at-triumf-dot-ca
Snail mail: 4004 Wesbrook Mall, TRIUMF, Vancouver, B.C., V6T 2A3, Canada

ATOM RSS1 RSS2