From: Dennis J. Hartmann (dhartma5@optonline.net)
Date: Thu Nov 10 2005 - 21:56:50 GMT-3
The current QoS Best Practices (SRND) is to tune all T-1 interfaces
to a tx-ring to 4 (even though they will default to 2 as Chris has pointed
out). Www.cisco.com/go/srnd
-Dennis Hartmann
-----Original Message-----
From: nobody@groupstudy.com [mailto:nobody@groupstudy.com] On Behalf Of
Chris Lewis
Sent: Wednesday, November 02, 2005 10:42 AM
To: vmashburn@fedex.com; ccielab@groupstudy.com
Subject: Re: tuning the tx-ring queue
Hi,
I don't know where that formula comes from or what it represents, perhaps if
you give the link, more comment could be provided. However, here is what I
usually say about the tx-ring.
the tx-ring is a FIFO queue providing buffering before the hardware line
driver, it allows an interface driver to maximise throughput. basically the
shorter the tx_ring, the shorter the potential worst-case delay that will be
experienced through that interface during congestion. If it is set too
short, the driver might not be able to maintain line rate dependent upon
the forwarding performance of the interface driver. Tx_ring tuning can
therefore be a trade-off between maximising throughput and minimising the
possible queuing delay incurred during interface congestion.
IOS self tunes the tx-ring based on the interface rate. If the line rate is
less than 2 meg, the tx-ring is 2, if it is lesss than 10 M, the tx-ring is
4, if it is less than 20 Meg, the tx-ring is 8 and so on.
The only time you would tune the tx-ring is if you think that the number of
packets in the tx-ring is too big for a given delay characteristic that you
are trying to design in to your network, as no re-ordering of packest is
possible within the tx-ring.
So if the tx-ring size is 4, and you calculate that 4 data packets ahead of
a voice packet for example, could blow your delay budget for that node
(based off interface rate), you need to reduce the size of the tx-ring.
So I'd say the only calculation you need is to add up the number of bits
that might be in the tx-ring, work out how long it takes the interface to
transmit those bits and decide if that fits in the delay budge you have for
that node.
Chris
vmashburn@fedex.com wrote:
Does anyone know how to interperate the formula for tuning the tx-ring
queue? Cisco's formula doesn't make any sense to me. They say to use the
following:
((Packet Size (bits) *8) * (tx-queue depth))/(interface bandwidth (bps))
However, I have a couple of issues with this formula.
1) What is meant by the tx-ring queue depth? ( I assume it means the current
size of the queue, but not sure).
2) The units of the result do not make sense. You have (bits(packet size) *
bits (queue size))/(bps(bandwidth)). The resulting unit is bit-seconds. I am
not aware of any physical unit that is measured in bit seconds.
Can anyone help out here?
Thanks
Vince
This archive was generated by hypermail 2.1.4 : Thu Dec 01 2005 - 09:12:06 GMT-3