About 4 days ago I started hearing the dreaded click of death on my one Windows computer. Problem is, I have two RAID arrays - one RAID1 with two drives and one RAID5 with four drives. Then also a standalone OS drive.

So I installed SMART monitoring software - which told me my OS drive is OK. I was about to reboot to check the Adaptec RAID BIOS for the other disks when I decided to try and find the problematic HDD by ear. I have a fan blowing on my machine as the heat levels are just insane. So I stopped the fan to listen carefully, but the noise went away. Took me 5 minutes to confirm that with the fan on, the click of death is present and with it off the click of death was absent.

Weird… This is an external fan. How can it influence a failing hard drive?

Review Part I here. The best I could muster was 21 minutes in a co-located setup. To make the comparison fair, I am going to express it as a percentage of the pre-optimisation application / PostgreSQL DB on production versus my development machine.

The performance increase between the Production App + Production DB server combo vs. my co-located development machine?

I recently purchased an OCZ Vertex 3 SSD 240GB to perform some PostgreSQL DB performance profiling on a large (80GB) database for one of my clients. I paid a premium of $600 for only 240GB purely because I wanted an SSD that does > 520GB/s read/write. To ensure I do indeed attain those speeds, I ensured my Gigabyte motherboard supported SATA3 at 6Gbps. It is no use connecting an SSD that does > 300MB/s to a SATA2 port, as SATA2 is limited to 3Gbps which translates to 300MB/s if you keep in mind 20% overhead is spent on the 8b/10b encoding scheme.

It follows that I was shocked to learn that I only managed to get 340MB/s read and 240MB/s write from the SSD connected to the onboard SATA3 port, with a Core i7 950 CPU. That is the same than an entry level SSD, which would have cost a third of the price!

After some troubleshooting I found this article. It turns out the onboard SATA3 ports on my Gigabyte X58A-UD3R are driven by a Marvell 9128 controller, which - to put it elegantly - sucks. This controller sits on a PCIe 1x bus as far as I gathered. PCIe 1x has a bandwidth of 5Gbps, which translates to 400MB/s. In real life, the effective transfer rates were much lower.

I recently had another instance where mobile phones (Android) refused to connect to an Exchange 2003 server. Using https://www.testexchangeconnectivity.com/, it stopped at

The help at Microsoft was misleading because I have already followed those steps. In this case, the issue was resolved by removing the Host Header Value from the Default Web Site in IIS: