In article <xoavlkgdyi09.fsf@sun.com>,
James Carlson <james.d.carlson@sun.com> wrote:

>Though that appears true for their implementation, other
>implementations will certainly vary. (Even other platforms from the
>same vendor may vary.)

yes, that's the point of talking about *implementation* limitations.


>I suppose it doesn't surprise me much that an embedded system has
>design limitations. Most general-purpose systems don't, though.

Perhaps most general-purpose systems explicit, documented limits on
the number of addresses they allow, but the classic BSD kernel
ifa_ifwith*addr*() functions tend to fall over if you configure a few
1000 IP aliases. I see that FreeBSD 5.4 still uses linear searching.

Long ago A UNIX vendor of my acquaintance hacked on the BSD net/if.c
to use hash tables instead of linear searching. That was before
HTTP included server host names and when an ISP customer or two
thought it better to serve 1000's of small customers with a few big
boxes instead of 1000's of small boxes. I thought it was a bad idea
and that customers who wanted such things should be told to to find a
UNIX system vendor dumb enough to make such changes for the supposed
market of giant boxes serving zillions of domains each with its own
IP address. No one listened to me before, during, or after the saga
of bugs and other problems with the new code. It's not a change that
is as easy as it sounds and it leaves permanent complications. I didn't
even get a real chance to say "I told you so" when the customer's
business proved fatally unprofitable for more fundamental but almost
as obvious reasons.


>I think the real question is: why did you ask?

agreed.


Vernon Schryver vjs@rhyolite.com