Quoting Gabriel G. Rosa ([email protected]): > I find the argument of journal overhead to be about as relevant in a > modern machine as the argument of software RAID overhead. That is to say, > not at all.
This line of thought isn't going to fly among system administrators, at least. ext2 has always been really valuable as a really fast operating system, ext3, even used with well selected journal options, is merely very good. In situations where performance matters -- and where a journal is not essential -- the choice matters. And even on my own servers, which are fundamentally bottlnecked on outbound bandwidth rather than disk, I'd rather not lose easy performance gains. > That is odd indeed ;) > > Can you elaborate a bit on this? Multiple extra layers of abstraction that don't, IMO, sufficiently repay that added complexity. Instead of just dealing in filesystems and their device names, you have a volume group on top of a partition, and a logical volume on top of that. More layers in the middle of your system to understand and manage (including device-mapper), and more to go wrong. Yes, you get LVM2 snapshots. I don't personally find that compelling enough. Your Mileage May Differ.<tm> > I think Bill's point is that swap spindle optimization is become largely > irrelevant with cheap and abundant RAM. Again, this is not a compelling argument for sysadmins, or anyone else who takes pride in getting easy gains of performance where they are available. > You can argue it's not a lot of extra work to set up, but it's also > not a lot of gains to be had over time. You have probably not seen systems thrashing for lack of it. > Until your storage is all solid state and seek times become > meaningless. Some of us (although not me yet) are already there. Indeed, one way to eliminate the need for competence at seek-optimisation is to eliminate seeking. ;-> _______________________________________________ vox-tech mailing list [email protected] http://lists.lugod.org/mailman/listinfo/vox-tech
