Topic: Server drops modem during HD Access

[David Boccabella]

I have a problem with my E-Smith 5.0 server.

Its is a Celeron 433 with 96M of memory using a dialup connection. The Drive is a 40Gig IDE and it also has a SCSI Tape drive attached

When I first set this system up I had 2 40 Gig HD's and was trying to do a software raide. But had to revert it back to a single drive for the following reason.

When the system is under load either whilst doing a backup verifit, or doing large file copies - the modem connection keeps dropping

Examining the logs gives me this

Apr 30 00:23:02 wepawet pppd[11313]: No response to 2 echo-requests
Apr 30 00:23:02 wepawet pppd[11313]: Serial link appears to be disconnected.
Apr 30 00:23:08 wepawet pppd[11313]: Connection terminated.
Apr 30 00:23:08 wepawet pppd[11313]: Connect time 1.6 minutes.
Apr 30 00:23:08 wepawet pppd[11313]: Sent 8857 bytes, received 5118 bytes.


When I had the raid installed it would drop and redial every 2 or so minutes.  Without the raid it is stable except when doing something that loads the HD side of things and then it will drop the connection every 10 mins or so.

Any suggestions PLEASE!!!..  As I get a dynamic IP address it is causing havoc with people looking at the site.

Many thanks..

Dave

[Charlie Brady]

David Boccabella wrote:

> When the system is under load either whilst doing a backup
> verifit, or doing large file copies - the modem connection
> keeps dropping
>
> Examining the logs gives me this
>
> Apr 30 00:23:02 wepawet pppd[11313]: No response to 2
> echo-requests
> Apr 30 00:23:02 wepawet pppd[11313]: Serial link appears to
> be disconnected.
> Apr 30 00:23:08 wepawet pppd[11313]: Connection terminated.
> Apr 30 00:23:08 wepawet pppd[11313]: Connect time 1.6 minutes.
> Apr 30 00:23:08 wepawet pppd[11313]: Sent 8857 bytes,
> received 5118 bytes.
>
> When I had the raid installed it would drop and redial every
> 2 or so minutes.  Without the raid it is stable except when
> doing something that loads the HD side of things and then it
> will drop the connection every 10 mins or so.
>
> Any suggestions PLEASE!!!..

Make sure that you have a good backup. Then do:

hdparm -u 1 /dev/hda
hdparm -u 1 /dev/hdc

That's drastically reduce the proportion of time when interrupts are disabled during hard disk accesses. The hdparm man page includes warnings about potential disk corruption, but I've never seen it. Be sure to let me know if you find otherwise.

Another option you could use is some of the other hdparm options, to enable DMA and 32bit I/O. There are also file system corruption possibilities with some of those options, and those I have seen.

Regards

Charlie

[littlalex]

Just a shot in the dark...but if you're using a "winmodem" with Linux drivers that might add to the problem. A "winmodem" as opposed to a "real" modem depends on the CPU and software to do the work that is done in hardware on the latter, sucking off precious CPU and mem resources.

[David Boccabella]

I havent replied for a few day as I have been monitoring the situation.

Many thanks Charlie..  It seems that tweeking HDPARM did the trick.
I disables the interrupts, and also added in the 32bit access as this is a fairly modern machine runing ATA100 drives.

I also moved the drives around so that the second HD was on its own IDE channel (a left over from the original software raid issues)

After doing all of that the system seems to be much stabler. I can even do backups and verifys without any problems..

Again thanks..

Dave

[robert]

gee David!

am I glad to find your posting - and the solution:
thanks Charlie.

Ive mucked with this for ages (looking for major-char-108 failure), then came across this : and viola, its fixed.

The box i had problem with is 1.7 Pentium 4 so should have had
plenty of performance...anyway FIXED IS FIXED woohoo

kind regards Robert Brocklebank

[robert]

Hi David!

May i ask what you did to enable ATA100 performance?
(my box shows ATA33 during boot)

kind regards Rob