Subject: Re: Network throughput and upgrading kernel questions
To: Henry R. Bent <firstname.lastname@example.org>
From: Blaz Antonic <email@example.com>
Date: 05/02/2003 11:53:09
> > When downloading something from external (outside local network) FTP i
> > got throughput of ~50 KB/s with Lynx and ~60 KB/s with wget;
> Interesting. I find that wget and lynx can usually get about the same overall
> speeds of 60KB/s here (512Kbit cable modem), and if anything lynx is a little
> faster. I take it you're using both lynx and wget from pkgsrc?
Perhaps i should mention that i am on 1 Mbit/s DSL line with maximum
throughput (of data coming in my direction, bandwidth is asymmetrical).
That results in ~120 KB/s on router, networked Linux box and networked
Windoze XP box; this (Windoze 3.1) box gets 16.6 KB/s max (i blame its
TCP stack) while Vax got less than 100% of 120 KB/s for locations
outside my local network too.Either way it doesn't seem to be able to
achieve same throughput as other boxes (i thought slow CPU was to blame)
but the difference between Lynx and wget really surprises me. I'd like
to hear about other people's experience too.
> > When downloading something from local FTP (another box on same network
> > segment) i got throughput of 150 KB/s with Lynx and 220 KB/s with wget.
> Huh. My 4k/200 is packed up at the moment (I'm in the process of moving) but I
> usually see local speeds of around 500KB/sec with both lynx and wget. Are you
> using the local disk or are you running over NFS? It's possible in either case
> that the disk interface or the overall system bandwidth are limiting transfer
> speeds; in any case it seems like you should be getting better speeds using
> either program. You don't happen to be using a local MFM disk? My experience
> with those drives is that they're pokingly slow.
Ouch :) I believe got got ~150 KB/s with Linux via NFS too, not too sure
about that tho (i was backing up my entire disk with MOPbooted Linux/Vax
before installing NetBSD, dd if=/dev/sda of=backup_location). The
figures i quoted are for storing data to local HD, which is SCSI RZ26L.
NetBSD doesn't use any NFS mounts and is totally indenpendent (except
that its controlling terminal, the Linux box is usually not up so i
telnet to it from my Windoze laptop).
I am using that safe writing thingy for FFS, if it matters (async write
? did i get it right ?), because of relatively frequent system freezes
when not using serial console (but socket connection apparently isn't
terminated so i don't think it comes to a HLT). I was under the
impression though that this only affects writing of multiple small files
(unpacking entire kernel source tree for ~50 architectures i don't need
Since i'm not very familiar with NetBSD yet and just about everything
seems to be disabled by default :) i haven't tried download in the
opposite dierction (from NetBSD box to some other box) yet.
> If you are using a local SCSI disk, is it possible you could benchmark with
> another drive? When I upgraded my DECstation 5000/120 from an RZ58 to a
> somewhat more modern Quantum drive (both ~1GB) I saw a decent performance
> upgrade even though both disks should be able to achieve the maximum thoroughput
> of the SCSI controller.
Unfortunately i have no SCSI drives in any of the x86 boxes and that
RZ26L is the only SCSI drive i got. These old SCSI-1 beasts are
extremely hard to come by locally (as are SCSI CD-ROMs, i don't have any
but i really wish i had one for my Vax) - they are either corrupted by
now or trashed or stored somewhere and nobody knows where since much
larger and faster drives can be purchased for relatively low price. I'm
reluctant to buy second hand mechanical items such as HDs or CD-ROMs
from eBay and similar sources, especially ones that are so old and can't
be tried out before the purchase.