Hacker News Reader

Comment by RyanShook | Hacker News Reader

One aspect sometimes overlooked is that the physical device degrades over time. Blocks of memory corrupt, connections erode, humidity or dust slows down cooling mechanisms. We tend to think of computers as either working or not but they can and do break down over time.

View on HN

Most of those effects you mention manifest themselves as obvious crashes.

If cooling is insufficient so as to cause thermal throttling, the effects are usually very obvious too, and not gradual. Neither will reinstalling software change that.

Most computer users never monitor their cooling effectiveness at all. They get used to the fans spinning up when the computer is doing anything that THEY think is non-trivial, even if it actually should be trivial. Then there's a smooth transition from "fans spin up but things perform normally" (the ailing thermal solution is able to hit targets without throttling) and "fans spin up and things perform awfully" (the cooling can no longer hit targets without throttling).

This can be explained just by thermal paste drying out and caking up, no more mystery needed. I've had 12 year old "workstation" laptops (the copper cooling alone was heavier than some entire laptops are these days) which started spinning up their fans for seemingly no reason. I dismantled them, cleaned off the old paste, put on a fresh paste, reassembled, and haven't heard the fans once since then. The cooling was good on day one, it got worse for unavoidable physical reasons, and it's easy to fix if you know how.

The problem is most users do not know this is happening and wouldn't know how to fix it even if they did. If they ask someone for help, 99% of the time they'll be told it's too old and they should upgrade.

Surely if a connection erodes it is either still connected and working or not connected and not working? It wouldn't get slower.

Systems are built to be fault tolerant because hardware errors are inevitable. Bit-flips in RAM and disks, line noise in transmission circuits, voltage fluctuations, etc. And these problems are statistical. In a new machine the chance of error may be 1%, then as it ages the error rate increases to 10%. Repeat something a billion times and those percentages mean you have a million errors. Not only in time but density; a chip with a thousand transistors is going to have fewer problems than one with 10 million. That's why a much older system can be faster than an old but more recent one. But also because the older system will be less fault tolerant. Back then the cost of handling a bit-flip or low-voltage wasn't higher so they ignored the errors and continued to run at full speed as it corrupted your data.

The connection itself wouldn't get slower, but the outer mechanism using that connection may retry, and the retry may be successful. The appearance to the user is one of slowness.

For example, one of the symptoms of networking packet is slowness. Assume 2% of packets are lost. A packet is lost, the receiver re-requests it, and the retry has a 98% chance of being successful. End result: the transmission completes, but at a slower pace.

The process of recognizing that a subset of a component has failed, recovering from the failure, and resuming would result in a slow-down. I'm not sure that's a full explanation for the experience though.

For something like an Ethernet connection a cable can certainly be half bad, good enough that you can transfer some data but with speed and reliability problems particularly involving weirder protocols that involve broadcast on the LAN and such.

Reply on news.ycombinator.com