Drobo Speed and the Mac Pro

I got a Drobo a little while back. Its a pretty cool device. I have however been a bit disappointed with the speed.

Searching through the internet I read that the FW800 interface on the Mac Pro ain't that hot, and that getting a PCI-E interface would prove faster. Well after hooping in newegg.com and $40.00 later, I tested it out.

Here are the results:

As you can see there is a pretty big difference. For example, sequential read with 256K blocks was about 38MB/s with the PCI-E card while the motherboard FW800 interface was about 22MB/s.

This benchmarking was done with XBench on OSX 10.6.1 with the Drobo having 4 Seagate 1.5TB SATA 300 drives.

Now the Drobo is faster, but still not as fast as I'd expect. Given that, Its still a pretty cool device, and speed is my only complaint.

Home Network Load Balancing

I'm been downloading some stuff with bit torrent via RSS feeds lately. With all the talk of throttling by ISPs, and to achieve domestic tranquility by having the MRS not bitch about slow internet at home, I've implemented load balancing with pfsense and my 2 broadband connections.

With bit torrent opening up an assload of independent connections to hosts all over the internet, you would figure its a prime candidate for load balancing. Well its not. I'm still trying to figure out why, but to achieve any sort of reasonable performance, I need to force it to a single interface with firewall rules.

Anyone with insight into why this is, please fill me in.

Netbook SSD Performance

I have a Dell Mini9. It has a 4GB STEC SSD drive. As SSD's are extremely different then HDDs in the way they operate, I was interested in its general performance under certain conditions.

My biggest interest is the difference between the stock Ubuntu 8.04 Hardy Heron and the New Ubuntu 9.04 Jaunty Jackalope. One of the primary differences between the two is that Jaunty supports EXT4 while Hardy supports EXT3. The kernel for Hardy is 2.6.24-SS LPIA, the kernel for Jaunty is 2.6.28-11 i386.

Now I know this is not a scientific test, as there are a bunch of variables that I have not controlled for, I.E.; kernel, kernel architecture, fs, and so on. I did use the noatime option on both ext3 and ext4.

All tests were performed with iometer. They were performed on /dev/sda1 as I didn't want to write directly to /dev/sda directly, as this would blow away the underlying OS. I may do that in the future.

The first set of tests is read performance, both random and sequential. I did this with 2 different transaction sizes, 4KB and 2MB. As the SSD block size is 4KB this is the smallest effective transaction possible. The 2MB size I thought was big enough to give an idea of larger transactions.

Both IO/s and MB/s are ascertained. The IO/s tells us how quickly a transaction can be accomplished. If you have multiple processes accessing different data simultaneously, the more IO/s you have the better. The MB/s is obviously how much data through-put that transaction provides.

Here are the results:

As you can see, Jaunty and EXT4 excel at every test.

Below are the write tests. To do the write test, the drive was aged. If you dont know why aging a SSD drive is important check out this article by anandtech.

Again, Jaunty and EXT4 excels. You can also see that the write through-put is very low. .03 MB/s. This is a result of the aging of the SSD. I will look at a fresh SSD's write performance soon.