Is this the word's poorest performing FreeNAS system?

[CAlbertson]

I think I must have the poorest performing FreeNAS system on Earth. I'm testing it as a backup server for Apple's Time Machine. It's been running now for about an hour and has backed up 8.41GB and says it has a total of 1.57TB to go. There estimate for completion is 13 days. That is TWO WEEKS.

Here is the setup. The Mac is a "Mid 2012 27" with 12GB RAM. I'm using the built in 1000BaseT Eithernet port, the WiFi is turned off.

The FreeNas system is not high end.

1) Intel Atom D525 CPU at 1.8GHz, 4GB RAM, The ethernet port (re0) is a Reltek even though this is an Intel motherboard.
2) I am using just one disk drive. A Seagate 7,200 rpm "barracuda". It is formatted as a UFS file system and shared using AFP.
3) I have a 10/100/1000 baseT switch with thee cables in it, one to each computer and a third going to my router. Color codes on the switch are as expected showing two 1000 connectins and the one going to the router as 100.

Any idea what to look for. This is a test setup and I don't have important data on it. I must have something misconfigured. What should I check?

 

[cyberjock]

I don't know a damn thing about Apple's Time Machine besides it does backups. Does it have compression? (My guess is yes) Is it set to maximum? It could be that you are CPU limited with Time Machine. 8.41GB over an hour comes out to just 140MB/min... or 2.3MB/sec! Whoo wee! That is smoking hot! Can I buy?

Also from what I've heard AFP isn't that great for speed, but I have no personal experience. Maybe you shouldn't have bought an Apple product(just kidding).

Is it possible you have a bad disk in either station? It could be that you are having trouble reading or writing the data.

 

[CAlbertson]

I thought that maybe the Mac had slow networking. So I connected a second Mac to the same switch and ran iperf. I get .93 gigabits per second between the two Macs using either one or five parallel streams. It seems to be a hard .93 as it does not change as I try different parameters in iperf. Then I get .4 gigabits from either Mac to the FreeNAS system. This is likely because of the RelTek ethernet chip on theFreNASS system. But even so 400,000 bits/sec is a usable performance level.

I thought Time Machine might be the problem so I opened up a terminal on the Mac and used "cp -a" to move a large 50 GB directory of (mostly larger) files. I'm still not getting even 2 megabytes per second. So, As poor as the Realtek controller is, it is still not the bottleneck.

I run "top" on the FreeNASS and on the Mac. The Atom processor on the FreeNAS is running about 77% idle with plenty of free RAM and on the Mac the i5 quad core is about 99% idle.

It looks to me like the network connection from Mac to FreeNAS is about 40Mbytes/sec but the data moves at 2Mbyes/sec

One more experiment: I replace the FreeNAS with a recent Macbook Pro. I simple share a directory under $home. With this setup I see better than 60Mbytes/second. The Macbook makes a pretty good files sever but then it is a 2.5GHz i5, not a 1.8GHz Atom. The acbook, while doing the file transfer is running with the (two core i5) about 70% idle

BTW, yes, I did read your guide. That is one reason I'm not even trying to run ZFS.

 

[bollar]

Netatalk isn't the most efficient module, but I was able to get read speeds reported in excess of 1Gbps (so in reality the theoretical maximum) on my higher-end Macs on both FreeBSD 8.3 and Netatalk 2.2.4 as well as OpenIndiana 151a5 and Netatalk 3.0.1. Writes were processor intensive and much slower. The test system had a Xeon processor and 16Gb/RAM, so it wasn't low-end.

As you know, TM isn't a good test, as it is a lower-level process that never uses more than a fraction of available processor and bandwidth.

Short of trying a new processor, you could try a couple of things on the server -- turn off atime and compression, first off. Then try reading from server to Mac -- I think that would be as good as it gets as far as reads go, without CPU and maybe RAM upgrades.

 

[jgreco]

Realtek is known to chew more CPU than a real ethernet chipset, and the Atom is a slow CPU. You are likely throwing away a lot of CPU cycles at ethernet processing. One thing to try would be to repeat your test doing raw file copies over AFP and then inspect what's going on with "top" on the FreeNAS CLI. 77% free sounds good but if it's real heavy on system or interrupt that may be skewing your results. The reality is that these systems have become complex enough that it is often hard to get things to run at 100% CPU for things that involve multiple complex subsystems (sigh).