tag:blogger.com,1999:blog-8489272546789521531.post2459104807445527420..comments2023-06-25T05:56:14.046-03:00Comments on Technical Musings: SSD Fade. It's real, and why you may not want SSDs for your ZILChristopherhttp://www.blogger.com/profile/15769619786580030432noreply@blogger.comBlogger5125tag:blogger.com,1999:blog-8489272546789521531.post-50821078296945677782012-12-15T11:46:53.056-04:002012-12-15T11:46:53.056-04:00Hi subdragon,
Since I use OCZ SSD's, I use t...Hi subdragon, <br /><br />Since I use OCZ SSD's, I use the OCZ Toolbox to perform my secure erase of the drives.Christopherhttps://www.blogger.com/profile/15769619786580030432noreply@blogger.comtag:blogger.com,1999:blog-8489272546789521531.post-486922055730153182012-11-25T17:46:17.893-04:002012-11-25T17:46:17.893-04:00hi Christopher
Would you tell me how do you "...hi Christopher<br />Would you tell me how do you "did a secure erase of the drive"<br />I try to format the ssd for zil, and re-enable zil, it seemed there is no obvious improvement.subdragonhttps://www.blogger.com/profile/12692212642811339593noreply@blogger.comtag:blogger.com,1999:blog-8489272546789521531.post-54408647313686309702012-01-08T17:51:47.310-04:002012-01-08T17:51:47.310-04:00The way I understand it, at most the metadata upda...The way I understand it, at most the metadata updates related to random async writes go through the zil (and I'm not even certain about that).<br /><br />Async writes can definitely be lost when the system loses power. Just think about it: the async write operation returns before the data has been committed to non-volatile storage. It logically follows that there is a time window for it to be lost before it does get committed.<br /><br />(Note that sync vs. async is orthogonal to sequential vs. random -- your reply seems to imply some confusion in that regard.)<br /><br />See e.g. here: http://constantin.glez.de/blog/2010/07/solaris-zfs-synchronous-writes-and-zil-explainedAndrasnoreply@blogger.comtag:blogger.com,1999:blog-8489272546789521531.post-44410244617819315642012-01-03T11:21:17.552-04:002012-01-03T11:21:17.552-04:00I will have to post on this again shortly, because...I will have to post on this again shortly, because there seems to be a lot of misconceptions about what the ZIL does.<br /><br />The statement I made about the ZIL is true for my pool, perhaps I shouldn't have said "your pool" but instead "my pool". <br /><br />However, I believe the only thing that bypasses the ZIL is a large sync write - that is sent directly to the disks to keep performance high. Otherwise, it all goes through the ZIL, including random writes, because these are normally waiting to be written in a transaction group, and we can't lose them if we lose power.Christopherhttps://www.blogger.com/profile/15769619786580030432noreply@blogger.comtag:blogger.com,1999:blog-8489272546789521531.post-31503637952335173532011-12-30T18:06:50.906-04:002011-12-30T18:06:50.906-04:00"Nearly everything written to your pool is wr..."Nearly everything written to your pool is written to the ZIL." -- from what I understand of ZFS, as well as based on my observations, this statement is wrong.<br /><br />As far as I am aware, the ZIL only matters for sync writes, few of which should normally take place on a system. Ordinary async writes bypass the ZIL completely.<br /><br />Of course, you with your VMs may get many sync writes (I don't know), but IMO then you should qualify this statement.Andrasnoreply@blogger.com