Disappointed

Jumbo frames

I finished a network upgrade today changing out my old Netgear gigabit switch for a SMC 8508T, and the old, flaky Netgear gigabit NIC for a D-Link DGE-530T, identical to the one in the resonant.org server. This is all 32-bit PCI, still, but that's all that my aging systems support. With jumbo frames now active from end to end, I'm still only seeing 28MB/s via HTTP (with about 98% CPU usage on Apache) and around 23MB/s (with about 40% CPU usage) via SMB. Since my uncached hard drive output on the server is only about 20MB/s, this isn't really that bad, per se, but I was hoping for much better. At this point, I appear to be capped by user-side CPU overhead and by the PCI bus itself, so I won't be getting much better until I completely overhaul both systems to have modern motherboards and processors. That's an order of magnitude more expensive than this experiment, so it's unlikely that I'll be doing it anytime soon.

On the plus side, the network instability appears to have cleared up. I still have one test left to run, but my first test case that reliably locked up the system within minutes has been running now for about half an hour without any problems whatsoever.

Unsubmitted Triangles

I had a hard time coming up with triangles to shoot for the most recent DPChallenge contest. My ideas for staged shots seemed trite and too easy, though a quick scan of the entries shows that nobody used the idea that I thought would get overdone: 4-sided dice. I guess I should have tried that idea after all.

The shot I ended up editing but then deciding not to submit was:

Triangles Within Triangles

Syndicate content