bauman@shell.com (E. G. Bauman) (07/03/90)
Greetings all. Just upgraded from a Sun 386i to a diskless Sparc 1+/GX with 24MB of ram. Now most of the time, this Sun just screams. However, when I'm using a moderate number (~12) of windows in Sunview, it'll sometimes page for a few seconds before even echoing what I type. This happens right after I switch focus to a new window. I'm paging over the ethernet to a 4/280 with Fujitsu 2382 drives on an SMD-4 controller. Is this normal? I'm sure there are worse things that could happen, but it's sort of embarrasing when I'm showing off the computer to have to wait for it to echo what I type. I guess a local SCSI disk for paging would help, but I'd like to avoid the extra noise in my office. All the above boxes are running 4.0.3. Thanks in advance for the advice. Evan G. Bauman Shell Development Company - Westhollow Research Center PO Box 1380; Houston, TX 77251-1380 bauman@shell.com or for the uucp types: {sun,bcm}!shell!bauman
graham@relay.eu.net (Graham Underwood) (07/05/90)
In article <9530@brazos.Rice.edu> you write: > >Greetings all. Just upgraded from a Sun 386i to a diskless Sparc 1+/GX >with 24MB of ram. Now most of the time, this Sun just screams. However, >when I'm using a moderate number (~12) of windows in Sunview, it'll >sometimes page for a few seconds before even echoing what I type. This could be due to you running out of pmegs (Page Map Entry Groups). The SS-1 has 128, which is less than other Sun4's, so runs out sooner. A pmeg entry is used up for each block of memory mapped up (e.g stack, code, file etc.) where a block can be from 4K to 256K. A program will therefore use many pmegs. When you run out of pmegs, the kernel will steal one from an inactive process - however there appears to be a nasty bug in the kernel, the effect of which is that it steals from an active process. This is particularly bad news if the robbed process is large because it gets swapped out. Since amount of memory mapped up by each pmeg can can be as little as 4K you will generally run out of pmegs long before you run out of physical memory. Under normal usage there would appear no be no way that the SS-1 could make use of more than 16 MBytes of memory. One benefit of this is that it saves on swap space. >All the above boxes are running 4.0.3. Thanks in advance for the advice. The bug is there in 4.1 too. Graham.