On Sat, Jun 2, 2012 at 2:50 PM, John R. Dennison jrd@gerdesas.com wrote:
On Sat, Jun 02, 2012 at 10:59:13AM -0400, Boris Epstein wrote:
A process implemented in the userland may not be as efficient as one implemented as part of the kernel - but that doesn't mean it can't scale well, does it?
Depends on ones definition of scale I suppose. I consider efficiency and performance one factor of scaling. To be completely honest about this I must admit that I've not spent a lot of time benchmarking any user space implementation in a large deployment but I wouldn't expect performance to ramp up based on scale.
I've always had a strong aversion to file systems implemented in user space versus kernel space as I've (personally) never found such an implementation that had what I considered good performance.
My needs, however, are not yours. If your requirements give you leeway for higher latency and slower overall performance perhaps a userland file system will work perfectly fine for you. As with all else in the IT sector use what works best for you :)
John-- Human beings hardly ever learn from the experience of others. They learn; when they do, which isn't often, on their own, the hard way.
-- Robert Heinlein (1907-1988), American science fiction writer, Time Enough for Love (1973)
John,
To be specific, I use UNFSD to export a MooseFS file system. MooseFS, by the way, is userland-process based too.
Be that as it may, I've seen situations where a comparably configured MooseFS client get to read at, say, 40 MB/s - which is fine - but the UNFSD at the same time reads at 40K/s(!) Why would that be? I mean, some degradation I can dig but 3 orders of magnitude? What is with this? Am I doing something wrong?
I can't believe it works the same way for everybody - who would use it if it did?
Boris.