I just finished re-formatting my Linux server. Previously I'd only used it mostly only for NAT to my other 2 computers. I got two more computers, and decided it would be very useful to use it for remote storage as well.

The comp specs are meagre (P1 @ 133mhz, 128mb ram)

I first tried installing RedHat 8, and then later Mandrake 9.1 which is what's on it now.

The "problem" is in the speed I get from it. I've got a 100mbit (switched) network, the other computers are running XP (and one will be running 98, but it's not set up yet).

I get an average of 2.5-3.1MB/s to and from the server, whereas I get +/- 6-8MB/s between two windows computers (which, granted have better stats, but the hard drives are comparable).

I thought it might have something to do with the mode (PIO vs DMA), but according to "hdparm" it's using DMA for both drives.

Next thought was maybe it was samba, so I tried FTP... I get very close to the same results with FTP (up and down) as through samba...

Then I thought maybe the computer just couldn't access the hard drives very fast and that was the bottleneck... but this would dictate otherwise:

-----
# hdparm -Tt /dev/hdc

/dev/hdc:
Timing buffer-cache reads: 128 MB in 2.93 seconds = 43.69 MB/sec
Timing buffered disk reads: 64 MB in 4.39 seconds = 14.58 MB/sec
------

I have a generic Realtek interface to the LAN... would upgrading the card to something "better" help, or is there anything else anyone can think of to cause the slowdown?

Thanks,
CG

Does ifconfig eth0 show lots of errors, overruns or collisions?

cheers

Jamie...

nope. none at all.

eth0 Link encap:Ethernet HWaddr 00:40:F4:45:88:10
inet addr:10.1.0.1 Bcast:10.1.0.255 Mask:255.255.255.0
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:16555260 errors:0 dropped:0 overruns:0 frame:0
TX packets:10080256 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:100
RX bytes:400845216 (382.2 Mb) TX bytes:4173838316 (3980.4 Mb)
Interrupt:11 Base address:0x4000

That all looks quite normal. Try keeping an eye on your CPU using 'top' while your are FTPing a large file accross just to check that the CPU isn't maxing out for some reason.

cheers

Jamie...

When FTPing, TOP reported ProFTPD using most of the cpu, with 0% idle

When trasferring via samba, smbd uses +/- 82%, with +/- 12% idle

I noticed python pops up intermittently and uses ~ 40%

Thanks for your help so far, BTW.

Are any NICs like winmodems in that they consume a lot of cpu to emulate missing hardware? The LAN card was a $3 card. As I recall it was a realtek (marketed under a different brand name)... I don't remember the model number. One of the other computers in my lan has a similar one, but that computer is ~1.4ghz.

-CG

Sounds like you might be experiencing a processor bottle neck when FTPing. Samba doesn't seem to consume all your CPU time which suggests there may be somthign else holding the system back. What type of network card is it exactly. I take it that being 100Mb that its a PCI card. What module are you using to drive the network card - 8139too - if its an TRL-8139 based card.

There are some differences in network cards/drivers under linux and I'd expect you to find you'd get better performance from a 3Com or a later Intel EEPro100.

How much memory have you got free? What does free tell you?

cheers

Jamie...

I'm pretty sure the module was 8139too... but I don't remember and don't know how to look it up. Yes, it is a PCI card. The machine has two cards in it and I know the other one was using tulip. It's a netgear card, and that's connected to the cable modem.

The font isn't a fixed width, so I'll just tell you the results

126424 total ram

Before transferring a file, I had 73.65MB in use, and 49.80MB free

While, and after (it didn't free up), transfering (using samba) I had 120.81MB in use, and 2.65MB free

-CG

I just pulled the case off and the chip on it is labeled RTL 8139C

Use lsmod
You can use [ code] and [/code] (without the space between [ and C) to force a fixed width font for things like that... Anyway it doesn't look like you are running out of memory but if you post the output from free then we an see how much swap you are using and whats going on with disk caches too.

cheres

Jamie...

NIC modules:
New values (closed notepad)


This is after the computer decided to randomly reboot itself... I think one of the DIMMs might have come loose because when it came back up it only registered 64mb... before noticing, I transferred a file, and actually got slightly higher results (~4MB/s instead of 3.5MB/s). Don't know if that was a fluke or what was going on there...

Looks like you've got plenty of memory available (90MB). Perhaps it is mainly a cpu or PCI bus limitation (or a combination of the two along with say the IDE controller or memory bandwidth), there doesn't appear to be a single item that's especially overloaded (the processor being the highest so far but no 100% when using samba).

It might be worth considering swapping the network cards around so the tulip based card is being used for your internal connection and seeing if that make any difference. I know the tulip chipset is quite well regarded and have plenty of drivers support, or isn't that a 100MB card?

cheers

Jamie...

Thanks for all of your help with troubleshooting... I'll probably just have to deal with it until I build another computer and move everything down one.

The reason I don't use the tulip card for the LAN is because it cuts out at high speeds (it used to be in one of the desktops, but the card would power down during large transfers... never for low bandwidth things like internet downloading or browsing), so this is the only place in my network it can go.

how hard would it be to switch eth1 and eth0? it works for a while, so I could at least see if a different type of card works better.

-CG

Just a case of changing the IP addresses around in Mandrake's config tool (whatever that is) and swapping the cables around as far as I can see.

Jamie...