On May 31, 2:56 am, James Kanze <
[email protected]> wrote:
[...]
Did you consider using Open Solaris on PCs?
I didn't have anything to do with the decision, but given the
body of existing, Solaris oriented code, it certainly should
have been a consideration (but probably wasn't). As it happens,
my part of the application is very I/O bound (it's responsible
for the transactional integrity of the entire system), which
means that switching to PC's is going to pose significant
performance problems for me---the CPU's may be as fast or faster
than the Sparcs we currently use, but the I/O bandwidth is
noticeably less.
I was advised to consider that a while ago rather than Linux.
I haven't really looked into it. So far I'm happy with Linux.
It probably depends on what you're doing. My experience is that
Linux is a good deal less reliable than Solaris (but that's
comparing Linux on PC's with Solaris on Sparcs), or even than
Windows. But it's probably reliable enough for a lot of things,
and the price is certainly acceptable. (I might also add that
most of the reliability problems I've actually encountered on
Linux seem to be linked with the X system. Which of course,
isn't a real problem for my server software.)
I note that it matters what distribution of Linux you use -
"How a programming error introduced profound security
vulnerabilities in millions of computer
systems."
http://www.technologyreview.com/Infotech/20801/page1
That sounds like a more general problem. If I understand the
article correctly, the software was originally counting on the
memory in some buffer not being initialized for some sort of
randomness. Which is a serious error to begin with. But of
course, if the desire is some sort of randomness, replacing the
initial lack of initialization with initialization with all
zeros is only going to make things worse. (The correct
solution, of course, would be to initialize with values read
from /dev/random.)