Re: NT speeds-ups?

Nils Myklebust wrote: > > Greg <greg@cyberramp.SPAMCHOKER.net> wrote: > snip..... > :My Recommendations: > :1. Set the FET_BUF_SIZE environment variable on both client (INet) and > :server. We found 8192 to be pretty optimal, though I suggest playing > :with it. The difference is HUGE!! The defaults are atrocious, and you > :send lots of mostly empty packets. We tested a couple of ODBC > :solutions, but they weren't even close. > > Could you please tell us what ODBC solutions you tested? > We are about to do this same thing (not with PB, but still) and for > many reasons prefer to use ODBC. Would be realy interested in some > input. > snip... Nils...sorry to break it to you, since I know you are a big Openlink fan, but that was the one we really tried to make work and couldn't. PB's native Informix drivers coupled with INet 5.01.WJ2 with FET_BUF_SIZE=8192 on both client and server just absolutely wiped the floor with Openlink. When we tried to get technical help from Openlink, on the 5% of the time they called back, they were totally lost. We dropped them from consideration pretty quickly after that. Not only was throughput less than half of INet, but Openlink chewed a lot of server CPU per session (5% ish of a six way Sparc 1000E if I recall correctly...). Too much for as many concurrent sessions as I was planning for. At the time, I really had high hopes for Openlink, and was quite disappointed. It should have been able to do much better. About that time we discovered Oracle was giving away it's INet equivalent free with its engine, so we torqued our salesrep into following suit, INet cost was no longer an issue. At this point, we're driving the available bandwidth at about 60% of it's theoretical limit, and 80ish or so is the best i've seen any client do (FTP specifically, INet not involved obviously....), so I doubt there is much room for improvement. Greg -- Please remove SPAMCHOKER. to get correct EMail address...thanks.