ZFS + Network Tunables + HP N40L

[VeGeTa-X]

Hi, I have a question about my tunables for zfs and network configs below and suggestions for better tunnables for better perfomance. I am noticing almost peaks and steep drops in network performance when transferring to my freenas server. I have noticed the network drops occur when I see the LED on my server that shows when data is being written to disk. I will also add hardware I am using any advice or tips would be greatly appreciated. Thx


HP ProLiant N40L Ultra Micro Tower Server
Kingston 4GB 240-Pin DD = 8GB
Port-Channel onboard nic and intel nic
4 x Hitachi GST Deskstar 7K3000 HDS723020BLA642 (0f12115) 2TB 7200 RPM
------------

vfs.zfs.arc_max 5074190464


kern.ipc.nmbclusters 10000


vfs.zfs.txg.write_limit_override 1073741824



vfs.zfs.vdev.max_pending 8



vfs.zfs.vdev.min_pending 2



kern.ipc.somaxconn 384



kern.ipc.nsfbufs 150000


net.inet.tcp.inflight.enable 1


net.inet.tcp.recvbuf_auto 1


net.inet.tcp.sendbuf_auto 1



net.inet.tcp.delayed_ack 0


net.inet.tcp.sendbuf_max 2097152


kern.ipc.maxsockbuf 2097152


net.inet.tcp.recvbuf_max 2097152

 

[VeGeTa-X]

Oh yeah forgot to mention I am using FreeNAS-8.2.0-RELEASE-p1-x64 (r11950)

 

[jgreco]

vfs.zfs.txg.write_limit_override 1073741824

I'm *pretty* sure your platform isn't capable of writing that much data per transaction group in a timely fashion.

Bug 1531 talks extensively about transaction group flushing and sizing of the txg. tl;dr: Don't make huge transaction groups, you get burstiness. The tuning described there works fine on 8.3.0-REL. Also note that it has nothing to do with iSCSI and is a general ZFS tuning issue, iSCSI just happens to be particularly sensitive to burstiness.

 

[VeGeTa-X]

I am using NFS for my shares, I noticed that one of my network cards on my freenas server is having the erratic issue the onboard network card and the extra card I have server has better network traffic. I will keep on investigating on this issue