Re: [gentoo-user] Btrfs: how to fix slow sequential write performance?

[William Kenworthy <billk@iinet.net.au>]

On 25/09/13 16:54, Marc Stürmer wrote:
> Greetings fellow Gentooistas,
> 
> at the moment I am doing some testing and evaluating of Btrfs on my own
> system. Why? Just because I can and I am curious about it.
> 
> Because I am doing that for evaluation purpose please don't give advise
> like "go ZFS" or something like that, thank you.
> 
> Kernel is 3.10.7, the file system is on one hard drive only (SATA).
> 
> And here comes my question: sequential write access seems to be slow as
> hell, actually something around like 10 Mbyte/s according to dd's output.
> 
> The same HDD performed under ext4 with around 90 Mbyte/s.
> 
> Command is something like: dd if=/dev/zero of=file.img bs=1G count=150
> 
> Nothing else is running at the time of writing, no scrub/balance on the
> file system or other process eating up much cpu time.
> 
> Mount option is "relatime" only. So is this a well known issue only with
> this kernel or generally with this file system at the moment?
> 
> Are there any possible fixes to squeeze better performance out of it or
> is this unlikely to happen so that I should better take that into
> account of my evaluation and maybe dump it then?
> 
> Thanks in advance.
> 

How old is the fs? - I am ran into a major slowdown due to COW snapshot
fragmentation (VMs on ceph using btrfs) - a known problem.  Using that
kernel is also giving me an eventual hard lockup if I try and do a
recursive defrag (on another system), but the ssd on an apple air
defraged fine.

btrfs is also seems very sensitive to how full it is.  Have you tried a
btrfsck? - some errors seem to cause slowdowns as well.

BillK