ZODB supports a client-server configuration. In that configuration, all of the CPU intensive work gets done on the clients. A single server can support a fairly high number of clients. It has an invalidation-based consistency model that allows clients to cache most of the data they read.
It's actually unlikely that the scalability limit you hit is with ZODB. In high-load Zope environment, you're more likely to hit a limit on the front-end. Generating dynamic content in Python can be fairly slow.
A lot of posters seem to have concluded that MIT sold out or that MS will corrupt the school. The worry seems to be that because MS provides the money, they will dictate how the money is spent, forcing MIT students or faculty to do something they wouldn't otherwise do. For example, MS would restrict MIT's ability to share the results of its research.
I think these worries are mostly misplaced. The amount of money is non-trivial, but a relatively small part of the MIT budget. MIT also has a long history of doing research with corporate money and freely distributing the results.
MIT's annual research budget is mostly likely more than $700 million. So a company spending $25 million over a few years probably doesn't get to dictate a lot. (It's a little mind-boggling to think, though, that Bill Gates's personal wealth could fund all the research for a good number of years.)
A good historical example to consider is Project Athena. MIT got millions of dollars from IBM, DEC, and Xerox to explore new uses of computing in the curriculum. The freely available fruits of that effort include X and Kerberos. It's worth noting that it presumably wasn't the companies pushing the free distribution of that software, because the Andrew window system, developed with funding from some of the same companies, withered because of licensing restrictions.
xFS is designed for machines that communicate over a fast network, i.e. a low-latency, switched network which will be faster than local disk. For example, the RAID file server on my local network (100Mbps switch Ethernet) consistently outperforms my local disk, although my disk isn't particularly speedy. The xFS paper (ACM TOCS Feb 1996) claims that local disk can be an order of magnitude slower than remote server memory.
xFS has a number of other advantages for locally distributed filesystems. In particular, it eliminates the need for a centralized server, which would otherwise limit performance scalability and reliability.
Of course, none of this matters much for the original poster. It sounds like he was *not* on a fast network with many other machines.
Slashdot Mirror
ZODB supports a client-server configuration. In that configuration, all of the CPU intensive work gets done on the clients. A single server can support a fairly high number of clients. It has an invalidation-based consistency model that allows clients to cache most of the data they read.
It's actually unlikely that the scalability limit you hit is with ZODB. In high-load Zope environment, you're more likely to hit a limit on the front-end. Generating dynamic content in Python can be fairly slow.
A lot of posters seem to have concluded that MIT sold out or that MS
will corrupt the school. The worry seems to be that because MS
provides the money, they will dictate how the money is spent, forcing
MIT students or faculty to do something they wouldn't otherwise do.
For example, MS would restrict MIT's ability to share the results of
its research.
I think these worries are mostly misplaced. The amount of money is
non-trivial, but a relatively small part of the MIT budget. MIT also
has a long history of doing research with corporate money and freely
distributing the results.
MIT's annual research budget is mostly likely more than $700 million.
So a company spending $25 million over a few years probably doesn't
get to dictate a lot. (It's a little mind-boggling to think, though,
that Bill Gates's personal wealth could fund all the research for a
good number of years.)
A good historical example to consider is Project Athena. MIT got
millions of dollars from IBM, DEC, and Xerox to explore new uses of
computing in the curriculum. The freely available fruits of that
effort include X and Kerberos. It's worth noting that it presumably
wasn't the companies pushing the free distribution of that software,
because the Andrew window system, developed with funding from some of
the same companies, withered because of licensing restrictions.
xFS has a number of other advantages for locally distributed filesystems. In particular, it eliminates the need for a centralized server, which would otherwise limit performance scalability and reliability.
Of course, none of this matters much for the original poster. It sounds like he was *not* on a fast network with many other machines.