T.R | Title | User | Personal Name | Date | Lines |
---|
1050.1 | | STAR::GAGNE | David Gagne - VMS/LAN Development | Fri Aug 06 1993 10:59 | 9 |
| The Ethernet does DMA on both transmit and receive.
The DEFZA does DMA only on receive. All transmitted packets have to be
copied to the DEFZA. This obviously costs CPU cycles. This could be
where you are losing your performance. The DEFTA (an FDDI Turbochannel
device) does DMA for both transmit and receive.
The version of VMS that supports the DEFTA is V5.5-2H4. The DEFTA is
not and will never be supported on VMS V6.0.
|
1050.2 | I don't think it's the DEFZA... | SCHOOL::WASHABAUGH | Born to be Mild | Fri Aug 06 1993 11:07 | 14 |
| > The version of VMS that supports the DEFTA is V5.5-2H4. The DEFTA is
> not and will never be supported on VMS V6.0.
I think you meant "The DEFZA is not and never will be supported on VMS V6.0"...
I don't think the problem is with the DEFZA. You are perfectly correct in
stating the the DEFZA does not have transmit DMA. However, I don't think this
would lead to the miserable performance that this person is reporting. I would
think that the problem would be some sort of network configuration problem,
maybe something that causes lots of dropped/retransmitted packets. However,
I am most ignorant in that area and cannot be of any assistence...
Doug
[Former fimrware engineer on the DEFZA]
|
1050.3 | | CSC32::B_GOODWIN | | Fri Aug 06 1993 11:22 | 7 |
| It may be that the DEFZA is transmitting so fast that the PC can't keep up and
so retransmissions must occur. If you are using DECnet as your transport for
Pathworks, I would look at some of the node counters under ncp and see if you
are retransmitting alot. You might be able to reduce this amount of data
retransmitted by lowering "pipeline quota" on the vax. The only other way to
see if this is happening is to put a analyzer on the network and analyze the
protocol.
|
1050.4 | | KONING::KONING | Paul Koning, A-13683 | Fri Aug 06 1993 11:53 | 4 |
| I remember something about the DEFTA not being supported on one of the VAX 4000
models? Correct? If so, was it the -90?
paul
|
1050.5 | ... | MUDDY::WATERS | | Fri Aug 06 1993 13:45 | 3 |
| VAXstation 4000 model 60 (a.k.a. PMariah) does not support DMA on the
TURBOchannel. So DEFZA/DEFTA won't work on VS 4000/60. 4000/90 supports
FDDI for sure, or you would not be reading these notes... See 417.2.
|
1050.6 | .1 is correct as it stands | STAR::GAGNE | David Gagne - VMS/LAN Development | Fri Aug 06 1993 15:27 | 5 |
| Regarding the comment in .2:
.1 says that the DEFTA is not and will never be supported in V6.0 and
that is correct. The DEFZA is supported in V6.0. The DEFTA will be
supported in a release after V6.0 and is now supported on V5.5-2H4.
|
1050.7 | Will keep you posted | BIGRED::SPARKS | I have just what you need | Tue Aug 10 1993 22:46 | 7 |
| I am going to re execute the tests myself, and will zero all counters
and make snapshots after each test. I still think the DEMPR would slow
the packets enough for the PC to handle them though, but it is worth a
shot. I will post any results. I am on vacation this week, so it will
be a few days, thanks for the replies so far.
Sparky
|
1050.8 | DEFTA seemed to fix it. | BIGRED::SPARKS | I have just what you need | Wed Aug 25 1993 10:39 | 8 |
| The customer upgraded to the DEFTA card and driver, did some system
tuning and the performance is back on track with no major peaks. I'm
not saying the problem was the DEFZA, but it went away when the card
did, also there was some system tuning done at the same time.
Thanks for the help.
Sparky
|
1050.9 | | BIGRED::SPARKS | I have just what you need | Thu Aug 26 1993 18:37 | 8 |
| I talked with the system manager some more, and they made a couple
other changes also. In the tests with the DEFZA they didn't have the
logical netbios$device logical defined, also the PC was loading the
last and lanses drivers. In the second test Last and Lanses were
disabled and the logical defined, so everything is somewhat
inconclusive. The performance now is very good though.
Sparky
|
1050.10 | No faster than Ethernet? | WARNUT::FORSHAWJ | Jon Forshaw | Fri Apr 15 1994 05:27 | 18 |
| I have a customer with PATHWORKS V4 and two Open VMS Alpha AXP 3000
Model 600 servers and about 1300 PCs. He has recently put his two
servers on an FDDI ring using a DEFTA-FA card with a Synoptics bridge
onto Ethernet segments with the PCs.
To his surprise he has seen no improvement in performance and has
asked is happening? Should he see an improvement in performance, how
can he demonstrate this to himself, is PATHWORKS a bottleneck and so
on?
I have not yet qualified the extent of how he is measuring the
performance and that he has everything configured correctly, but I
thought I would drop a note in here to see if I could get some initial
reactions to this situation. Any comments gratefully received.
Regards,
Jon.
|
1050.11 | Bridge adds latency? | SSDEVO::PARRIS | RAID-5 vs. RAID-1: n+1 << 2n, in $$$ | Fri Apr 15 1994 14:43 | 10 |
| Re: .10
I'd expect that the conditions under which you'd see a noticeable improvement
would be if the Ethernet was saturated before the move to FDDI, and that by
dividing up the clients across multiple Ethernet segments you eliminated that
bottleneck.
Moving the servers from Ethernet to FDDI does mean you're introducing
additional latency in the bridge. How fast can the Synoptics bridge filter and
forward between FDDI and Ethernet?
|