Audience::All
UUID: c078319b-b156-404f-a48b-1e639dd734b6
Earlier today, in the midst of an ongoing project, I noticed that
- The temperature of a single physical CPU was running at 104° F; about 10° hotter than expected.
- There were a large number of Firefox tabs open (40-odd), as is typical when abnormal high temps are seen.
Knee-jerk reactions seldom lead to any good outcome.
Step one was to take a quick a quick shot at verifying the problem. Since Firefox 35, we have been able to set privacy.trackingprotection.enabled=true in about:config. I had done that the day before, (before the problem was noticed) but had not restarted Firefox. This time I bookmarked all pages, restarted Firefox, and reloaded all tabs. Temps returned to normal. Though based on a single datum, I may be able to assign a provisional cause. Go, me! Possible progress! I did some ancillary things, such as noting before and after memory usage (in case the kernel scheduler was part of the problem), etc.
None of that really mattered, though. In the greater scheme of things, it seems likely to be irrelevant. At the very least, a lot more open research is needed.
Temperature
First off, the widely-taught inverse correlation between temperature and lifetime, may be entirely bogus over large domains, and seems highly likely to be far more nuanced than is often taught.
Perhaps it matters in, say, applications related to RF power systems, such as radars and electronic countermeasures, but I haven't worked in those fields in years. Though messing up fire-control radars was tons-o-fun. I care a lot more about, to use an overly-generic term, IT.
HPC centers, the hyper-scale service providers, and large enterprises, all care about bills due to power. Supply costs, conversion efficiency, what is devoted to heat dissipation, thermal effects on the longevity of vast fleets of servers, etc.
Google does not provide OpenSource code at anything like the rate that they consume it, but they do provide landmark papers, which is at least partial compensation. Failure Trends in a Large Disk Drive Population (2007) was such a paper, and it implied that increased temperatures enhanced longevity.
Temperature Management in Data Centers: Why Some (Might) Like It Hot (SIGMETRICS12, University of Toronto) extended those results to DRAM, set some boundary conditions, etc.
In the same year (2012) No Evidence of Correlation: Field failures and Traditional Reliability Engineering was published, but I have not digested that yet. It's corporate, and I've only recently discovered it. I'm interested in the intersection of security and traditional reliability engineering (it's the 'A' in the CIA security triad, after all) you might want to read it as well.
Obviously, this is nothing like a comprehensive literature search. But I really doubt that simplistic schemes purporting to draw an obvious inverse correlation have any merit.
Tracking
Unfortunately, this post has gone on for too long. Not in terms of what should be covered, but in what I have time to cover. It's 1915, there are still Things To Do, and it is already going to be a late night.
Some things are going to be left for a possible future post. I tend to want to leave this sort of thing to more consumer-oriented security sites, where 'Don't Run Adobe Flash' might possibly help someone. An obvious problem is that many of the consumer sites do not cover tracking issues, and some of those that do are either biased or intentionally misleading. That sucks, but it isn't as if I am going to write a definitive post, complete with an economic history, this evening.
No comments:
Post a Comment
Comments on posts older than 60 days go into a moderation queue. It keeps out a lot of blog spam.
I really want to be quick about approving real comments in the moderation queue. When I think I won't manage that, I will turn moderation off, and sweep up the mess as soon as possible.
If you find comments that look like blog spam, they likely are. As always, be careful of what you click on. I may have had moderation off, and not yet swept up the mess.