0004219666@MCIMAIL.COM (Bob Stine) (03/01/91)
A client of mine is developing a real-time, distributed system that will use the
TCP/IP protocol suite for control & monitoring. A component in the system will
serve as a gateway between a LAN and a T1 link to (another gateway on) another
LAN.
In addition to control traffic, we will be performing down-line loads to
components on the distant LAN, sometihing like:
LAN T1 LAN
Server ------ GW --------- GW --------- Destination
1 2
Now, we don't want the bulk traffic to cause overly-long delays to the real-time
traffic. The control traffic will all be on a single 56kbs channel of the T1,
framed in HDLC. We were contemplating giving the down-line load traffic its own
channel, and using an application at GW2 to receive and distribute big files to
the destinations on the far LAN.
We will be writing our own device driver for the T1.
To me, a "natural" solution to this problem seems to be to use separate IP
addresses for the control and bulk-data channels.
Questions:
1. Would this require separate drivers to get any performance improvements
(i.e., would a sinlge HDLC device driver become a bottleneck)?
2. Is there a better approach?
3. What are the hidden "gotchas"?
4. Where could I look for more info?
Thanks,
Bob Stine
bstine@MCIMail.com
c_rstine@hns.comkasten@EUROPA.CLEARPOINT.COM (Frank Kastenholz) (03/02/91)
> From tcp-ip-RELAY@NIC.DDN.MIL Fri Mar 1 10:21:04 1991 > From: Bob Stine <0004219666@mcimail.com> > To: tcp-ip <tcp-ip@nic.ddn.mil> > Subject: IP addressing & T1 channels > > A client of mine is developing a real-time, distributed system that will use the > TCP/IP protocol suite for control & monitoring. A component in the system will > serve as a gateway between a LAN and a T1 link to (another gateway on) another > LAN. > > In addition to control traffic, we will be performing down-line loads to > components on the distant LAN, sometihing like: > > > LAN T1 LAN > Server ------ GW --------- GW --------- Destination > 1 2 > > Now, we don't want the bulk traffic to cause overly-long delays to the real-time > traffic. The control traffic will all be on a single 56kbs channel of the T1, > framed in HDLC. We were contemplating giving the down-line load traffic its own > channel, and using an application at GW2 to receive and distribute big files to > the destinations on the far LAN. > > We will be writing our own device driver for the T1. > > To me, a "natural" solution to this problem seems to be to use separate IP > addresses for the control and bulk-data channels. > > Questions: > > 1. Would this require separate drivers to get any performance improvements > (i.e., would a sinlge HDLC device driver become a bottleneck)? > > 2. Is there a better approach? > > 3. What are the hidden "gotchas"? > > 4. Where could I look for more info? > > Thanks, > > Bob Stine > bstine@MCIMail.com > c_rstine@hns.com Bob, I suggest that you use type of service routing for differentiating the traffic "streams". Do something like assigning the download traffic "high throughput" and the control traffic "low delay". It sounds like you are building your own router/gateway box so you can have it do the right thing. Differentiating things by IP address would be "a bad thing". First, IP addresses just identify an endpoint of the communications -- they are not meant to identify type of service. Second, what if you decide to change IP addresses? You might want to take a look at things like the OSPF spec and some of the Internet Drafts and RFCs on things like policy routing and the like. Hope this helps Frank Kastenholz Clearpoint Research Corp.