On Sat, Aug 09, 2008 at 02:04:28PM +0200, Wim Vandeputte mailinglist only wrote: > > On Sat, Aug 09, 2008 at 12:53:23PM +0200, Bernd Walter wrote: > > You can easily get a CRC read error for a logical block far away from > > the block which was written during power fail. > > So a rw partition can harm a ro partition - at least for power failures, > > which I've seen quite often in practice with all vendors I'd used. > > This is one more reason why I always advise people to buy industrial > grade flash > > http://www.siliconsystems.com/technology/powerArmor.aspx > > You also get a 6 bit ECC (compared to a 2 bit ECC), comprehensive wear > leveling and elimition of bit-flip errors.
I only saw CRC problems together with power loss. ECC or better ECC can't help here. > The downside is that it's a more expensive solution. Where I live power loss is very unlikely to happen. But I have customers which power-cycle the system for every kind of problem. For example I had a customer running Soekris boxes for VPN to use special software on a VMS system. They powercycled the Soekris boxes several times until they noticed that just the software license of their software on their VMS run out. > You can industrial grade flash from Sandisk (like the 5000 series), > Transcend and Silicon Systems, the only problem is comparing the > performance between vendors, but so far from 5 years experience > with Sandisk industrial grade (SDCFB-xxx-201-80 and 5000 series) and > Silicon Systems, I have not seen a single failure, no data loss > or corruption. > > > I don't think wearing out a media is a real risk today, but power > > loss in the wrong moment is still a problem and with growing Flash > > block sizes the risk is growing. > > Yes, I've seen that as wel with consumer grade flash. This is quite interesting, because I was never interested in extended wear leveling, temperature range or shok resistance. Those factors are already quite good for consumer grade cards. I never thought that the power loss problem is solved with industrial cards, because the underlying problem is a principal one, but I can imagine that this can be worked around by clever design. -- B.Walter <[EMAIL PROTECTED]> http://www.bwct.de Modbus/TCP Ethernet I/O Baugruppen, ARM basierte FreeBSD Rechner uvm. _______________________________________________ Soekris-tech mailing list [email protected] http://lists.soekris.com/mailman/listinfo/soekris-tech
