> This thread may have outlasted its relevance to the list,
> but I'd like to know how ebay built their architecture.
>
> Do you have a good url on what they looked like before the
> crash and after?
News articles about their crash say that they run on a single large
Sun server, which has had not-so-great uptime. The big crash came when
an OS bug scribbled on their (RAID) disks. Whoops. Their idea of fixing
their reliability problems is to buy another big Sun server and use
Sun's HA product.
The reason that a cluster would work well for them is that all of
their auctions are independent, so that's an embarrassingly parallel
situation. Searches of all auctions for keywords etc is not
embarrassingly parallel, but that's only a fraction of their
processing, and the results of such searches can be allowed to be
stale in some aspects.
Most of the high-end websites run on Linux boxes are exploiting some
sort of embarrassingly parallel aspect of a system. For example, all
of the complicated cgi scripts on Yahoo are run on a big farm. There
are databases behind most of those scripts, but most of the processing
is in the interface, and that bit is embarrassingly parallel...
-- greg