Physical Disk writes affect other networked machines

[Nathanial Thelen]

I have 4 modern Compaq servers networked via a Gigabit ethernet
switch.

1 is a DL380 and has 3 148GB drives RAID 5 (since 270GB logical
drive). Disks are completely full (1 GB free)

3 are DL360

The 3 are reading lots of data off of the Drive server (80Mbit/sec)
then broadcasting out to internet through second NIC.

I noticed that every once in a while (can't pin down a regular
interval but is in the ball park of every 5-10 minutes) all the
servers stop sending any data for 4-5 seconds. This isn't just the
data they are streaming off of the drive server, but all network
traffic stops (can't even ping them on the external NIC).

I opened up the performance monitor on the drive server and during
this 4-5 seconds the PhysicalDisk counter: "Disk Writes/sec" goes WAY
up and all Reads go down to 0. The writes are usually next to nothing
since the server is only serving out.

So I copied some of the files from the drive server to one of the
other servers and had all the servers access those files from that
one. When the main drive server does it's mega write, even the files
coming from the secondary drive server stop sending so it isn't just
that the drive is too busy writing to read.

The networking is all standard windows file sharing. The Switch is a
HP Procurve. The wiring is all copper.

Tried isolating the process that is doing the writing via the Process
counter in the Performance monitor, but none of them seem to change
much when this is happening.

Done lots of searching and talking to people about this, no luck.

Any ideas?

Thanks,
Nate

 

[Robert Bogue [MVP]]

To state the obvious ... You've got current network drivers and hard disk
drivers on all the servers, right?