Banita Liviu wrote: > I am just trying to see if I got the concept right regarding RX/TX > descriptors. I am having some problems with a highly loaded server ( 4 oracle > DBs on top ). It's a Solaris 10 138888-08, and i know these are the > OpenSolaris forums but bare with me, because i don't think it makes a > difference in this specific case.
There've been substantial changes between S10 and OpenSolaris, both in the drivers themselves and in the way the stack operates. You really should be going through Sun's support group for this. > Now help me understand if I got this right. A hardware interrupt to read the > ring buffer, will only occur if at least 80 TX/RX descriptors are ready ? Not necessarily. The normal way such hardware operates is that it forces an interrupt immediately if it hits the limit on the number of receive descriptors, but will generate interrupts on a delay if at least one descriptor is ready. In other words, the "only if" should not be true here. If it were, then the driver would be fairly useless; you'd have to sit around waiting (possibly forever) for another 79 packets to come in every time you got one. It seems logical that this mechanism is related to the symptoms you're reporting, but I suspect there's more to it. > If yes, why on same types of server but with less network load the "initial hang" doesn't occur ? > > If i got this all wrong please tell me. I am just putting head to head some bits and pieces of informations I got from around. Work with support. That's what they're there for. -- James Carlson 42.703N 71.076W <[email protected]> _______________________________________________ networking-discuss mailing list [email protected]
