[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference noted::decnis

Title: DEC Network Integration Server (DECNIS)
Notice:Please read note 1 to use this conference effectively
Moderator:MARVIN::WELCH
Created:Wed Sep 18 1991
Last Modified:Thu Jun 05 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:3660
Total number of notes:15082

3630.0. "How many LLC2 relay supported by DECNIS" by HGOM19::MARKWU () Fri May 09 1997 00:36

    Dear colleagues:
    
    	Could anyone tell me that how many LLC2 relay links supported by
    DECNIS ?
    
    	I have a custom who have 30 X.25 switch needs use account, 
    	I wonder how many DECNIS600/500 and UNIX host they need ?
    
        Thanks.
  
    
    
    Best/regards
    Mark
T.RTitleUserPersonal
Name
DateLines
3630.1More details required about the configCOMICS::WEIRJohn Weir, UK Country SupportFri May 09 1997 05:0975
Mark,

Please think about your questions -- unless you just want to keep
"wondering" with nobody bothering to answer.

If you just want to know how many LLC2 links, then read the SPD, which
will tell you how many DTEs of different sorts and how many SVCs.

From my experience, the most critical issue is the performance, which,
simplistically, can be divided into:

	1. Packet rate (X25L3 packets per second transiting the box)

	2. Call rate (calls being relayed through the box)

The packet rate for an MPC-I or MPC-II is something around 300 pps. For an
MPC-III it is somewhere around 1000 pps (I think I saw the figure 1100
somewhere).

If you have relatively lengthy sessions -- eg users log on via X.29 at the
start of the day and log off in the evening -- then you can forget the
call rate. But, if you have short transactions, such as credit card or
ATM (Automatic Teller Machine) transactions then the call rate is highly
critical, as you may only be doing a few packets for every call and you
might want hundreds of calls per second.

Be warned! I have just had a Customer problem where the Customer did have
short transactions and could get no more than 4.5 transactions per second
before the DECnis started to drop connections. I did a load test with an
MPC-I (same as MPC-II) to see how many calls/sec  I could get through the
box (no data packets at all -- just a call and a clear). The MPC-I was
saturated at 7.5 calls/sec with V3.1-11. This version is expected to be as
good as any other. This implies that an MPC-III might saturate at around
25 calls per second (but I do not yet have one to test).

I did the same load test on a RouteAbout Central EI, and achieved a rate of
22.5 calls per second, but that was limited by my test rig and not by the
RouteAbout. (ie the Alpha 3000-300 making the calls was 100% CPU bound and
the RouteAbout was still not having any problems.) I have no idea how fast
the RouteAbout Central EI can go -- maybe more than 50 calls per second ??
All I can say is that it did not seem to be trying very hard when it
achieved 22.5 calls/second.

In conclusion, if you want an X.25 Relay, then you have a number of options.
There are very limited performance figures, so if you really want to know
the comparitive performance then you will have to run benchmarks.
(Alternatively, fund me to run the benchmarks ;-)) My gut feel is:

	- The DECnis is primarily a very good DECnet/OSI/IP Router, and a
	  good bridge. X.25 performance is almost an afterthought. If you
	  don't want routing or bridging then the DECnis is an expensive,
	  low performance, choice for X.25. (But, it does work well as it,
	  and its predecessors, have been used for many years and the bugs
	  have been shaken out...)

	- A Digital UNIX box running WAN Services (X.25) is a better price
	  performance option than the DECnis. Also, as there are many different
	  Alpha models, so you can match requirements to cost. This software
	  has been around in production quality for a year or two. (Early
	  versions were very buggy, and fixes were slow to come, but the
	  quality of current versions is pretty good.)

	- RouteAbout Central or Access are new to the market, so expect to
	  find bugs ... It looks as though the price performance will be
	  several times better than a DECnis or UNIX box, but nobody knows ;-)

If you want any more than this sort of general advice, then you are going to
have to think about your questions and provide a descriptions of what you
want to do and the required configuration...

Regards,

	John

3630.2MPC-III X.25 Relay calls/sec performanceBUNTY::WEIRJohn Weir, UK Country SupportTue May 20 1997 05:0139
	Just in case anyone interested in DECnis X.25 performance stumbles
	across this note, I thought that I'd add the results of my MPC-III
	testing of X.25 LAN/WAN Relay calls per second.

	Based on my earlier MPC-I testing (7.5 calls/sec at saturation) and
	the relative CPU power of an MPC-I (300pps) and MPC-III (1100pps), I
	was hoping to achieve about 28 calls/sec with an MPC-III processor.

	The actual measured performance was somewhat less than this -- The
	MPC-III achieved a peak of 24 Relay calls/sec. Measurements were
	consistently in the range 20 to 24 calls/sec with an average of just
	over 22 calls/sec. Be warned, that if you overload the DECnis the
	performance appears to drop off. I am not sure whether this is real
	behaviour of the DECnis or just a side effect of the way I was
	testing.

	For completeness, the test setup was...

	As I only have low power Alpha systems (DEC 3000-300) to drive the
	tests, I made each call traverse the DECnis 4 times. In this way
	an Alpha call rate of 1 call/sec would generate 4 relay calls/sec.
	I assumed that LAN/WAN performance was the same as WAN/LAN, so the
	results are the average of these.

	The four passes through the DECnis were:

	1. Alpha VMS LLC2 to loopback connector on WAN port

	2. loopback connector on WAN port to LLC2 on UNIX Alpha as a relay

	3. LLC2 back from UNIX relay to DECnis and a different WAN port

	4. Loopback connector on 2nd WAN port to LLC2 and back to 1st Alpha

	Regards,

		John

3630.324 calls/s does seems to be the limit...MARVIN::WELCHTue May 27 1997 07:1718
Thanks for posting your results. I also expected nearer 28 calls/s and only 
got ~24 in testing. However, I was using the SVC tester software and I thought
this was perhaps the limiting factor. The software is written in MACRO and 
creates SVC connections between a RX and TX process. The code now runs on 
OpenVMS Alpha as well as VAX.

I only had 1 Alpha and 1 VAX and even connecting them back-to-back couldn't 
get any significant increase in the call rate. Also the rate drops off as more
SVC's are established, both in the back-to-back operation and with the DECNIS,
so the 24 calls/s was measured over the first 100 calls.

There are no plans to try and improve the call rate and at least it is better 
than the MPC-I/II figures. 

Steve.