On 05/02/2011 13:03, Gary Gendel wrote: > I tend to agree with Matt. > > I've had several catastrophic power incidence on my server running > OpenSolaris without incident (such as a UPS gone haywire). Since ZFS > does transaction-safe reads and writes, the consistency on the disk > has an extremely low probability of getting corrupt. But there are > always two things to consider... Powering off the server this way > could cause the disks themselves to write garbage and ZFS (or any > other filesystem) won't have the opportunity to fix it. Also, there > may be pending things in the ZFS cache which won't be written. You > can disable the ZFS cache, but that will impact performance. > > If it's a situation where you don't have control over the servers > (someone turns off the master switch), then I truly suggest that you > put a UPS on the server and have a daemon (nut, apsupcd, etc.) that > get's the signal that the power has been killed and shuts down the > servers gracefully. > > Based upon the goals of Hammer, it looks like ZFS covers them all. It > looks like, on Hammer, you have to run a a prune/reblock on the > filesystem nightly to conserve resources. I would assume that heavily > pounding during the day will limit the effective utilization greatly. > How much overhead do you need for the prune/reblock, 10%, 20%? What > happens if there isn't enough room to run it? > > You're skirting on very thin ice if you pull the plug to shut down the > servers no matter what system you use. Interestingly, there are no > specs for Hammer in the FS comparison chart. This doesn't sound like > a seasoned production filesystem which would scare me even more. > > On 2/5/11 6:35 AM, Matt Wilby wrote: >> On 05/02/2011 11:28, Basil Kurian wrote: >> >>> Here we are running backup servers only during business hours (web >>> development firm). At the end of day , Sometimes the servers are >>> powered >>> off , without issuing any commands to shutdown them cleanly. >>> >>> On 5 February 2011 16:45, Matt Wilby<[email protected]> >>> wrote: >>> >>> >>>> On 05/02/2011 10:53, Basil Kurian wrote: >>>> >>>>> Hi all >>>>> >>>>> >>>>> Currently in our company we are having a Dragonfly BSD backup >>>>> server with >>>>> Hammer filesystem in it. One senior system administrator in our >>>>> compay >>>>> >>>> chose >>>> >>>>> Hammer filesystem some times back, because Hammer filesystem can >>>>> easily >>>>> recover in times of power failures. I 'm compelling him to move to >>>>> ZFS >>>>> >>>> (on >>>> >>>>> FreeBSD or OpenIndiana). How durable is data on an ZFS pool (RAID or >>>>> >>>> mirror) >>>> >>>>> when power failure happen frequently. >>>>> >>>>> >>>> Hi, >>>> >>>> I use ZFS on my backup servers. I can't say we've ever had power >>>> failures because they're in a datacenter, but I would invest in a >>>> decent >>>> UPS if it's a regular occurrence. Hard drives don't generally like >>>> having power suddenly removed..... >>>> >>>> ZFS on OpenIndiana will give you nice features like file system >>>> compression, dedupe, snapshots and constant checksumming of of in-use >>>> blocks. >>>> >>>> I've never used Hammer before, but HammerFS doesn't appear to offer >>>> compression yet. >>>> >>>> Just my opinion, but I'd say ZFS is more widely used than Hammer. >>>> >>>> Matt >>>> >>>> >>>> >>>> _______________________________________________ >>>> OpenIndiana-discuss mailing list >>>> [email protected] >>>> http://openindiana.org/mailman/listinfo/openindiana-discuss >>>> >>>> >>> >>> >> I don't understand why you would ever want to shut a server down in this >> way. >> >> Can't you just setup a cron to shut them down at a certain time? >> >> >> _______________________________________________ >> OpenIndiana-discuss mailing list >> [email protected] >> http://openindiana.org/mailman/listinfo/openindiana-discuss >> > > > _______________________________________________ > OpenIndiana-discuss mailing list > [email protected] > http://openindiana.org/mailman/listinfo/openindiana-discuss >
One other thing to consider if you are going to use Hammer is that it appears to require a cron job to be run to free up space/blocks in order to recover space on the file system. To quote the manual - "The system cron automatically runs HAMMER cleanup functions at least once a day." So obviously make sure you move the cron job to run whenever the server is up. Don't know how this impacts operational performance. I would feel uneasy about relying on a cron job to perform essential file system tasks. _______________________________________________ OpenIndiana-discuss mailing list [email protected] http://openindiana.org/mailman/listinfo/openindiana-discuss
