News for hackers. Stuff that matters.
Re:Three years later, Mr. Moore is still letting us punt on ... by kyledrake on Friday January 27, 2012 @02:31PM (#3519746)
The one warning I want to provide though is that not all SSDs are created equal: Make sure you get one that writes its cache to the disk on power failure, or you're going to be in a world of hurt.
Re:Three years later, Mr. Moore is still letting us punt on ... by krupan on Friday January 27, 2012 @03:57PM (#3520121)
Can you explain more? You aren't talking about writing data that is cached in the SSD to a regular hard-drive, are you?
Re:Three years later, Mr. Moore is still letting us punt on ... by tesseract on Friday January 27, 2012 @05:10PM (#3520415)
I think he's talking about writing data in the SSD's DRAM cache into the actual flash.
Re:Three years later, Mr. Moore is still letting us punt on ... by krupan on Sunday January 29, 2012 @04:16PM (#3526288)
That's what I thought too. Fusion-io drives all flush their write buffer to the nand flash on a power cut event, I wasn't aware that some SSD's didn't. People use iodrives as a caching layer in conjunction with tradition spinning hard drives, so wondered if possibly he was referring to that kind of setup.
Re:Three years later, Mr. Moore is still letting us punt on ... by xolox on Saturday January 28, 2012 @06:50AM (#3521882)
To guarantee on-disk consistency, programs like database servers call fsync for every transaction. By definition an fsync involves events on the hardware level and even in an SSD this gives a slowdown. The SSD can use a write cache to speed this up. However should you lose power and the write cache is not backed by a small battery, you lose your most recent write(s) even though the latest fsync call guaranteed that the changes were on disk. The battery allows the SSD to successfully flush the write cache to disk in the event that the computer shuts down.