centuriond
Dabbler
- Joined
- Nov 27, 2012
- Messages
- 12
Hey everyone,
I typically get Samba transfers of 60 - 80MBps on my setup with 8Gb (see signature for the rest); however, I've noticed that my write speeds are similar to that of Atoms. Drives are set to AHCI in Sata 3 (6Gb/s) with default compression settings on the pool. The pool is divided into 2 datasets (1x 20Gb jail and the remainder 10.7Tb used for the main media). I've seen transfers hit 110MB/s and stay for some time until dropping down to the aforementioned range. I am also aware of Samba's single threaded design but my cpu is clocked at 3.4Ghz and the cpu is hardly being worked in the reporting.
I've set autotune to set the variables as needed, but as I monitor my transfers, I see that the ARC never comes close to using the rest of the free ram (never drops below 3.4GB free) during transfers where I'm transferring a small number of large files.
Here are the autogenerated ARC and Kmem sizes:
vfs.zfs.arc_max = 4016353536
vm.kmem_size = 4462615040
vm.kmem_size_max = 5578268800
And some modified values I found that I had used when attempting to improve write speed before:
kern.ipc.maxsockbuf, net.inet.tcp.recvbuf_max, and net.inet.tcp.sendbuf_max are all set to 16777216
Given that I have 3.4Gb free, shouldn't the kmem_size_max and thus zfs.arc_max be significantly larger?
Thank you for any assistance and/or advice!
I typically get Samba transfers of 60 - 80MBps on my setup with 8Gb (see signature for the rest); however, I've noticed that my write speeds are similar to that of Atoms. Drives are set to AHCI in Sata 3 (6Gb/s) with default compression settings on the pool. The pool is divided into 2 datasets (1x 20Gb jail and the remainder 10.7Tb used for the main media). I've seen transfers hit 110MB/s and stay for some time until dropping down to the aforementioned range. I am also aware of Samba's single threaded design but my cpu is clocked at 3.4Ghz and the cpu is hardly being worked in the reporting.
I've set autotune to set the variables as needed, but as I monitor my transfers, I see that the ARC never comes close to using the rest of the free ram (never drops below 3.4GB free) during transfers where I'm transferring a small number of large files.
Here are the autogenerated ARC and Kmem sizes:
vfs.zfs.arc_max = 4016353536
vm.kmem_size = 4462615040
vm.kmem_size_max = 5578268800
And some modified values I found that I had used when attempting to improve write speed before:
kern.ipc.maxsockbuf, net.inet.tcp.recvbuf_max, and net.inet.tcp.sendbuf_max are all set to 16777216
Given that I have 3.4Gb free, shouldn't the kmem_size_max and thus zfs.arc_max be significantly larger?
Thank you for any assistance and/or advice!
