| |||||||||||||
The Linux Storage and Filesystem Summit, day 1The Linux Storage and Filesystem Summit, day 1Posted Aug 9, 2010 16:10 UTC (Mon) by nix (subscriber, #2304)Parent article: The 2010 Linux Storage and Filesystem Summit, day 1
a Google developer said that the feature is attractive for internal useIt's attractive for any application that has a heap of >2Mb (for the portions of the heap before the end), which is most of them. It'll probably even speed swapping: swapping in enforced 2Mb chunks is much more suited to modern disks than swapping in tiny little 4K pieces.
(Log in to post comments)
The Linux Storage and Filesystem Summit, day 1 Posted Aug 13, 2010 11:18 UTC (Fri) by i3839 (guest, #31386) [Link]
That's not generally true for SSDs, especially if NCQ is supported. When swapping in latency is usually more important than throughput, so you would want to only swap-in what you need. Maximizing thoughput by reading stuff you don't need is slightly silly.
With write-out 8 channels * 4 NAND chips per channel * 4KB page size = 128KB should be enough to get max write throughput (less for SLC). Buf if you're swapping, you're generally also swapping in, and writing a lot data will kill read latency, so you might want to limit the writes anyway.
|
|||||||||||||