>>>>> "W" == W Phillip Moore <[EMAIL PROTECTED]> writes:
>>>>> "Rens" == Rens Troost <[EMAIL PROTECTED]> writes:
Rens> I hope that the pts changes will make it do what I need it to;
Rens> We've been waiting for R/O access out of quorum since
Rens> 1993. Without R/O access out of quorum, AFS is really
Rens> unsuitable for widely distributed production environments.
W> Perhaps not, if you are trying to create one large AFS cell,
W> spread across several sites segmented by low-bandwidth WAN links.
Exactly.
W> This is a limitation of AFS we accepted as a deisgn constraint,
W> and we have been able to work around it. AFS can be used to
W> support a " widely distributed production environment", and we
W> are well on our way to making AFS the primary production
W> distributed filesystem in one of the worlds largest Investment
W> Banks.
With one of the worlds larger technical staffs. :)
I read the paper you did with Xev. Aurora is good work; You did the
right thing for the environment you have, a webwork of large LAN
clusters ("buildings") that is connected close to to 100% of the
time. You had to scratch-build a system ("VMS") for cross-cell volume
synchronization. I've considered the same hting here.
If AFS shipped with such a system, it would be a __much__ better
solution for a much broader audience. Or if there were a good
third-party one.
In contrast to the traditional big enterprise model you have at MS, my
environment has many much smaller LAN clusters; hedge fund outfits
with only a couple of users in each "building" all needing to access a
global namespace. I could serve these people with AFS on a
cell-per-campus if I had good enough (any?) multi-cell synchronization
tools, but that might get prohibitively expensive.
I'll go out on a limb and say that as cheap wide area connectivity
proliferates, big networks will look more and more like mine and less
like the traditional 80's big-shop ethernet.
The bottom line is that I cannot deploy AFS outside of my main office
until I can get a distributed filesystem that is resilient to WAN
outages. I was promised that this would be AFS 3.4. What I am hearing
now is makig me nervous.
If I need to staff up for a project as ambitious as aurora to make
AFS work reliably work in my network, then I guess AFS is not for me.
This would make me sad, though, since I was specifically promised that
this issue (continued read-only access to volume space during network
partitions) would be resolved to my satisfaction in the 3.4 release by
transarc. "Sad" in this context would mean "no longer a customer".
AFS is good stuff in the traditional big-shop with several large,
tightly concentrated campuses. Not all shops are like that.
-Rens
