Strange results from IE8 test locations

[ctext.org]

I was really pleased to see a Singapore host had been added - luckily performance for our site from Singapore looks pretty good.

However, I noticed a strange issue in the waterfall output from the Singapore IE8 host: the CSS and JS files often (but not always) display "0ms (Request canceled)", and then appear to be immediately re-requested. This doesn't seem to impact performance at all - which is odd in itself, since it seems to indicate that the browser canceled and then immediately re-issued the request, since "0ms" is certainly less than a RTT latency.
http://www.webpagetest.org/result/101107_AYTD/

It turns out that the Dulles IE8 host also exhibits this behavior - hadn't noticed it before since it doesn't actually seem to impact performance:
http://www.webpagetest.org/result/101107_AYSB/

Server logs only seem to show these files being requested once for these page views.

Anyone have any ideas what's going on?

[pmeenan]

My theory is that that is how IE8 pre-caches DNS (though it does look really strange when the real requests come right after). I have seen it pretty frequently with IE8.

-Pat

[crm911]

The Singapore test server seems to have gone. Hope to see it back soon.

[pmeenan]

Sorry, every now and then you will see agents disappear if they haven't checked in in a while (and we get alerts so we can work with the partners to get them back online). Any outages are usually pretty short and I'll send out a notice if we're expecting a location to be down for an extended period.

Thanks,

-Pat