I started out getting IO errors and certain files went missing, the system was 
useless.  When I rebooted the box, the SSD disappeared completely off the SATA 
bus.  This was after 24/7 constant IO for a month.

The device was a SuperTalent "commercial industrial temp" 32GB SSD. I sent it 
back and I was hoping to get a newer model (the ewiz.com line-up changed 
between order and return.)  But I got the same junk model back from them.

The controller on this thing sucks rocks, with random I/O per second count like 
a hard disk.  (I thought high end SSDs were rated in the tens of thousands of 
random IOs per second these days?)

Anyways I'm not suggesting that flashrd should be important on an SSD.  I 
certainly wouldn't want to use a tool like flashrd on an SSD, it was really 
intended for low-end flash.  If you spend $300 or $600 on an SSD, it should 
last like a hard disk, that's the idea.

As a side note, I've deployed several used Juniper M series routers with 
RE-333s for SONET/TDM termination to IP.  These old REs are Pentium II 333 SBC 
that run back end routing protocols and the CLI.  They come with CF and a 
laptop hard disk.  Juniper leaves both writeable, flash writes are somewhat 
frequent with logs, atime, configs, and so on.  The first RE-333 I got was 5 
years old (this was in 2004) and the flash was completely toast.  You have to 
take the SBC apart and replace the CF, reload JunOS, etc.  Since then I've gone 
through about five more RE-333s in more Juniper chassis and almost every time 
the flash has IO errors or won't even boot.  I still have boxes with 32MB CF 
running OpenBSD (as RO) that work fine.  Juniper's newer equipment (EX3200 
switch for instance) comes with an internal USB key now.

Obvoiusly there are lots of different approaches to these problems... I am 
trying to maximize cheap flash for use with cheap boxes like PC 
Engines/Soekris.  Juniper gives you redundant engines so when one fails you can 
just return it, under contract of course.  Except for all the EOL hardware 
that's still running :)

Kevin Chadwick [ma1l1i...@yahoo.co.uk] wrote:
> On Thu, 15 Jul 2010 19:59:00 -0700
> Chris Cappuccio <ch...@nmedia.net> wrote:
> 
> > I continue to kill R/W flash (last year, I killed a brand new SuperTalent 
> > server-class SLC SSD after 1 month of use, testing some huge and scary Java 
> > NMS app, jffnms or something like that. This app is an extreme example, 
> > while monitoring 1500 devices, it kept the SSD maxed out much of the time)
> 
> Interesting, what size was/were the flash disk(s) that failed? 
> 
> Were you getting write errors or did it die completely?

-- 
I know nothing except the fact of my ignorance -Socrates

Reply via email to