Messages in this thread |  | | | Date | Mon, 27 Jul 1998 18:06:30 +1200 | | From | Chris Wedgwood <> | | Subject | Re: 2.1.111 VM and low memory machines |
| |
On Mon, Jul 27, 1998 at 01:41:00AM -0400, Albert D. Cahalan wrote:
> Except for some broken old Ethernet cards, your data seems to > indicate that the default NFS block size should vary with the > total amount of memory.
Nah, I think it should default to 8k. For those with low memory (i.e. below 16MB) I think its reasonable that they have to change their mounts arguments to make things more sane.
> There is a TCP/IP window size CONFIG_* option that might deserve > similar treatment.
No there isn't. If you mean CONFIG_SKB_LARGE (which, I think, used to allow large TCP windows), it not used anywhere in the kernel source and should be removed from the Config.in file(s). I've submitted patches in the past that included removing this.
If you want large windows in 2.1.x, then try to convince DaveM and whomever else has a say in this that it doesn't break stuff. I have sysctl patches that allow up to 64k windows (without scaling) which I find work great over satellite, but I'm told windows over 32k break buggy stacks.
Arguably, TCP stacks should use wscale, but not that many stacks seem to support this (basically, only Linux 2.1.x and windows 95+/98, not NT, not Solaris).
I have to say, I've never seen it break anything, but I've not looked all that hard either. I'm also of the opinion, if it breaks some obscure stacks then too bad. Negotiated tcp options (timestamps, wscale, SACK, etc) break Ultrix - but do we disable them by default?
-cw
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.altern.org/andrebalsa/doc/lkml-faq.html
|  |