sob@harvisr.harvard.edu (Scott Bradner) (08/09/90)
Well it's that time of year again. We are getting ready to do another
run of network device tests. We plan to test four types of devices.
a/ Ethernet to Ethernet local routers
b/ Ethernet to Ethernet local bridges
c/ Ethernet to xx routers
d/ Ethernet to xx bridges
where "xx" = 4 & 16MB token ring, FDDI, serial link, etc
We plan to test as many Ethernet to Ethernet local routers as we can get
before Interop. We will post the results just before Interop.
After Interop, we will then test the other devices.
For the Ethernet to xx type devices we want to do a simple (for us) setup:
----- -----
==========| D |xxxxxxxxxxxxx| D |==========
----- -----
where ==== is an ethernet
xxxx is the other type network
We would then test between the Ethernets. i.e. we are not ready to
generate FDDI & count FDDI packets. This may or may not yield useful
information where "xxx" = FDDI.
The tests will include as many protocols as we can do, at least TCP/IP
and DECNET, with IPX & AppleTalk Phase II on the short list.
We want to do the range of tests that the IETF Benchmarking Methodology
Working Group is recommending. This includes:
device throughput at various packet sizes
a graph of output vs input from 0 pps to max legit pps
for packet size on network
effect of 1% error packets on throughput
effect of various filtering options on throughput
device latency
bidirectional throughput
(send equal # of packets through the device from
each side)
throughput with multi channels of data through multi-port device
There are other tests that the WG is looking at that are not yet defined
enough for us to do yet. If you want to join the WG mailing list send a
request to "bmwg-request@harvisr.harvard.edu".
We expect to get some of the tests done before Interop the rest after.
Now here is why this is being posted:
a/ If you build a bridge or router that you are willing to have tested (abused?)
please send me a note. Since we will only be doing the Ethernet
to Ethernet local routers before Interop, please hold off on the
notes about the rest until after Interop.
The testing will start the week of Aug 20th.
We will want the router(s) { we are willing to do more than one
type per vendor } for a few days. The ideal is to have a vendor
person come and set things up (so we don't get it wrong somehow)
{also I don't want to read that much documentation all at once}
NOTE: All test results on this first batch will be made public,
after Interop we may be willing to do tests on prototype devices
where the results are private, but ONLY for prototype devices.
We do want to encourage the development cycle but not to encourage
vendors suppressing information.
b/ If you have an opinion on the tests described please let me know
what it is (There may be no way to do the test that you would
like, but it is useful to know if we are off the mark.
If there is an interest in another testing & results talk at Interop like
I did last year & someone manages to reserve a space, I'll be glad to get
something ready.
PLEASE send mail directly to me (sob@harvard.edu). I'm going off for
a week in sunny New Orleans for SHARE (the IBM user's group) and the
newsgroup(s) will have expired by the time I get back. (not enough disk
space). Do not expect any reply until I get back.
----- please don't call, I will not be in for a week and a half
and don't want my voicemail to overflow.
Scott Bradner
Dan Lanciani