> I made a 1.3.89 kernel that fixes
>
> - I had missed one place where we should have called tcp_send_delayed_ack().
This fixes the ftp hang problem for me. As far as throughput is concerned,
NNTP and TELNET are now at a good level. Unfortunately between 1.3.84 and
1.3.89 I've seen another 20% slowdown on sustained PPP throughput:
1.3.8[0123] 1.3.84 1.3.86 1.3.86+ 1.3.89
---------------------------------------------------------------
NNTP ~3.5Kb/s 3.3Kb/s 3.6Kb/s 4.9Kb/s 5.1Kb/s
Compressible FTP ~6.0Kb/s 6.4Kb/s 6.1Kb/s 5.1Kb/s 5.0Kb/s
I appreciate that NNTP transfers will vary according to the type of
postings and number of articles. All of the above are a "slurp" of about
20 groups with about 2 days worth of postings (2,500 articles ?) which
under Windoze I can usually transfer at between 5.0 and 5.5Kb/s
I haven't had time to tcpdump the ftp transfer to see where the slowdown
is coming from. I'll try that ASAP.
> - various drivers updated (this includes the scsi reset handling, although
> only the BusLogic driver really uses that currently)
> - RTC driver - Linux/SMP needs it
> - SVGA bootup minor bugfix
> - "struct page" updates for the alpha too
> - MD and loopback shouldn't deadlock any more (confim, please?)
I would really like to see the MD code fixed "correctly" I've been meaning
to check out the MD code myself, but how many other projects am I involved
in at the moment ???
> - some of the more obscure network drivers work again
> - sound driver ioctl bugfix
> - ELF core-dumps more complete
> - iso9660 bugfix (probably the one that caused some problems for people with
> some CD's..)
Hmmm... Now where was that "buggered" CD ?
> - SMP should compile cleanly again
> - verify_area() bugfix (this was mentioned on the kernel list already: if you
> did that patch by hand you have to reverse it to apply the 1.3.89 patch)
>
> Thanks for trying this out (you ARE going to try it out, aren't you??),
Of course Master - us masochists like restoring backups when we don't read
a patch carefuly enough :-)
> Linus
Steve