It’s been a couple of weeks since I built a home NAS using a HP Microserver N36L with 8GB RAM, FreeNAS 8.0.2-RELEASE and 4 x 2TB Samsung F4 hard drives configured as a RAIDZ2. Apart from a scary incident which resulted in an unexpected real world test of RAIDZ2 resilience, the NAS has been pretty stable although I’ve not been blown away by read/write performance over the network. I didn’t really want to get into fine tuning ZFS this early as I was hoping the out-of-the-box performance would be good enough, but it looks like I’m going to have to do a bit of investigation to understand why performance is not as good as I had hoped.
It’s worth mentioning that I was also experiencing regular incidents of the NAS dropping off the network and reappearing several seconds later. This was particularly noticeable when SSHing onto the box using Putty, only to have the shell stop responding and the connection terminated a few seconds later. At the same time the web GUI would also stop responding and any remote file shares would also disappear.
Checking the FreeNAS logs didn’t show anything scary such as disk problems, so I Googled a bit and found many reports of problems with the on-board Broadcom based NC107i embedded network controller on the HP Microserver N36L. Users report regular network disconnection and reconnection problems and many have resorted to installing a separate quality NIC (such as an Intel PRO/1000 server or desktop card) in one of the PCIe slots. This sounded promising and I was all set to order a NIC when it dawned on me that I had been playing about with configuring my various network devices for jumbo frames support and when I couldn’t get it to work reliably had forgotten to revert my Win XP PCs NIC settings back to a default MTU of 1500! As soon as I did this the NAS network connection was steady again so I’ve delayed the purchase of a separate NIC… for now at least!
Testing network speed with iperf / jperf
Given the numerous reports of problems with the on-board NIC in the N36L, the first test I wanted to perform was a low level network test using iperf and its GUI front-end jperf. Luckily iperf is bundled with FreeNAS so it was simply a case of starting it in server mode using the command:
Then I fired up jperf on my iMac and ran a few basic tests…
The results were very positive! After several runs the average TCP transfer rate was around 910 Mb/s (or around 113 MB/s) which must be near the theoretical maximum throughput for a Gigabit network. Now these were not exhaustive tests for any long period or under sustained load, but the on-board network controller appears to be doing its job at least some of the time so I don’t think that’s the main cause of poor performance.
So next I think I need to start drilling down into testing the raw hard drive IO performance and then maybe onto a bit of ZFS tuning. But that will have to wait until another post 🙂