A few photos
I still do plan on editing and posting those shots from last year's trip to Peru. Someday. :)
An update to my previous post - I've found that the Linux device mapper (and therefore LVM) causes a major performance hit. dd from my RAID-5 array (/dev/md0) directly gives 210MB/s throughput. dd from a device mapper 'linear' device layered directly over /dev/md0 gives 80MB/s. I know benchmarks can be fickle, but this is a major difference. Benchmarking LVM on my workstation at work (only one hard drive) shows a major (although smaller) hit as well. Unfortunately, this is not an April Fool's joke. If performance is vital, Linux LVM is probably not the thing to use right now.




1 Comments:
I'm not sure it's safe to say that using LVM2 causes a major performance hit. Generally, LVM2 has no significant impact on performance. Doing dm-linear on top of software raid is a pretty uncommon configuration. So, there's that. Sometimes on faster SANs you may see an issue with request splitting (dm splits everything down into a large number of small requests), but usually that is masked by IO scheduler or fs tuning. I wouldn't jump to blame device-mapper on an un-scientific benchmark - there are many major deployments out there that use LVM2...
Post a Comment
<< Home