On 05/23/11 9:54 AM, Gordon Messmer wrote:
> On 05/23/2011 07:23 AM, Michael Schumacher wrote:
>> As far as I understand, SSD are fast at reading and slow at writing.
> A good SSD will be substantially faster at writes than a disk drive, as
> well. Because there's no head seeking around a platter, latency is
> vastly better, which provides a massive performance advantage in many
> (if not most) use patterns.
yes,butt.... SSD has to erase and write a LARGE block all at once, so
they don't do so well with the sorts of 8k random writes that write
intensive applications like relational databases commonly perform. To
write a single 8K block would require reading the whole flash block
(something like 64k or 256k), flash erasing that block, then rewriting
the whole thing. This would be painfully slow. So, what the drives
do instead is remap blocks rather randomly onto 'new' space, caching
them in drive-local buffer memory, then flashing whole blocks at once.
once the new space is used up, performance tends to degrade as they have
to scavenge scattered free blocks.
_______________________________________________
CentOS mailing list
CentOS@xxxxxxxxxx
http://lists.centos.org/mailman/listinfo/centos
