Performance issues with Alpha and VAX on same CI?

Performance issues with Alpha and VAX on same CI?

Post by Scott Viet » Tue, 27 Jan 1998 04:00:00



Here's a quick techie question to take your mind off Compaq's recent
purchase of DEC...

Has anyone heard of performance degradation caused by having both
VAX 6000s and DEC 7000s on the same CI bus?  I was talking with
another system dude and he mentioned that his cluster (which has both
VAX 6000 and Alpha 7000 systems) runs "slower" than if he only had the
7000s
and the HSJ42s on that CI.

I've never heard of a VAX 6000 "slowing down" the CI bus?  Is this true?

If so, please elaborate on why the 6000s slow things down.

Thanks.  :^)

-Scott Vieth
Sr. Systems Engineer
Detroit Medical Center

 
 
 

Performance issues with Alpha and VAX on same CI?

Post by Gotfryd Smoli » Wed, 28 Jan 1998 04:00:00


[cut]
+Has anyone heard of performance degradation caused by having both
+VAX 6000s and DEC 7000s on the same CI bus?  I was talking with
+another system dude and he mentioned that his cluster (which has both
+VAX 6000 and Alpha 7000 systems) runs "slower" than if he only had the
+7000s and the HSJ42s on that CI.
+
+I've never heard of a VAX 6000 "slowing down" the CI bus?  Is this true?
+
+If so, please elaborate on why the 6000s slow things down.

 Don't know exactly. Also don't know if HSJ42s is a dual - then
my message can be superfluous.
 But AFAIK *if* you have dual pathes then is important for
efficiency where pathes you use... Have check ?

 Regards - Gotfryd

--
=====================================================================
$ ON F$ERROR("LANGUAGE","ENGLISH","IN_MESSAGE").GT.F$ERROR("NORMAL") -
                THEN EXCUSE/OBJECT=ME

=====================================================================

 
 
 

Performance issues with Alpha and VAX on same CI?

Post by Hoff Hoffm » Thu, 29 Jan 1998 04:00:00


:Has anyone heard of performance degradation caused by having both
:VAX 6000s and DEC 7000s on the same CI bus?  I was talking with
:another system dude and he mentioned that his cluster (which has both
:VAX 6000 and Alpha 7000 systems) runs "slower" than if he only had the
:7000s and the HSJ42s on that CI.

  You might want to ask what this other "system dude" was refering to.
  I can only guess what he meant.  :-)

:I've never heard of a VAX 6000 "slowing down" the CI bus?  Is this true?

  There are VAX 6000 systems ranging from excruitatingly slow to just
  plain slow -- example systems include the 2.8 VUPS or so for the VAX
  6000-210, and 13.0 VUPS or so for the VAX 6000-510.

:If so, please elaborate on why the 6000s slow things down.

  Slower systems tend to hold onto locks and tend to block "critical
  paths" longer -- I ran one of the classic "odd couple" VMScluster
  configurations in years past, a VAX-11/750 (circa 0.6 VUPs) and a
  VAX 8800 (circa 6 VUPS per CPU).  When the VAX-11/750 accessed a
  common file or a common data structure, particularly something that
  the VAX 8800 needed to access, the VAX 8800 would sit-and-spinlock
  until the slower VAX-11/750 (slowly) finished with it.  One could
  notice when the VAX-11/750 started heavily accessing (and locking)
  SYSUAF, for instance.

  This behaviour will obviously become more noticable among systems with
  shared resources and rather larger performance disparities, and when a
  faster system waits for a lock served by or held by the slower system.

  I would tend to use the LOCKDIRWT SYSGEN parameter to "encourage" locks
  and resource mastership to occur on the faster node(s)...

        --

  As for "slowing down the CI", the mere presense of another CI adapter
  on the CI has little or no impact on CI performance.

  Discussions of the (very small) overhead involved with polling additional
  CI adapters will be ignored, as will all discussions of the dual-path vs
  single-path CI adapters, as I would guess that neither of these were what
  the "system dude" was refering to.  :-)

 -------------------------- pure personal opinion ---------------------------

  note to those folks not contributing spam -- there is no ZZ in my address

 
 
 

1. Alpha network performance issues over leased line

Hello,

We have a strange network performance problem.  We have a 2Mb leased line
terminated at each end by a Cisco 2600 router.   A Digital Alpha DS20 is
present in both locations, along with a network of NT boxes. One Alpha is
loaded with VMS 7.1-2 and the other with 7.2-1.  The network at either end
is Switched ethernet.

So we have

| NT boxes|   |DS20 (7.2-1)|
____|___________|______      Location A
                   |
      |Baystack 350 |
      |10/100 switch |

    |Cisco 2651 router|
                  |
                  |
                  |    (2Mb link)
                  |
                  |
    |Cisco 2621 router|

      |Cisco Catalyst|
      |  3500 switch |
_________|____________    Location B
         |                      |
| NT boxes|   |DS20 (7.1-2)|

All boxes can connect over IP, but performance varies dramatically depending
on the destination box.

In summary, performance declines dramatically when the Alpha is the
recipient of the data, and is OK otherwise. The alpha in location B has only
recently moved there.  They used to talk quite happily to each other before
this, when the second Alpha was in another location again.  The only things
that have changed are the routers at each end (used to be Nortel ARN
routers) and the seond Alpha now plugs into a Cisco switch rather than a
Nortel Baystack 350 it plugged into before.

There appears to be no congestion on the link.  Can anyone think of any
reason why the choice of router (Cisco) would have such an effect, but only
on traffic to the Alphas? Or should I be looking at something else?  Some
measurements on performance are given below.  .

- An FTP put from the DS20 in A to the DS20 in B is dreadful at about 25kB/s
- An FTP put from the DS20 in B to the DS20 in A is also not good at about
50kB/s
- An FTP put from any NT box to any NT box (over the link) is  about 170kB/s
in both directions
- An FTP put from an NT box in A to the DS20 in B is again about 25kB/s
- An FTP get from an NT box on one side of the link to a DS20 on the other
is about 170kB/s
- Local performance from NT boxes to Alpha's (both PUT and GET) seems OK at
about 4MB/s in both locations

Any help greatly appreciated!!

2. Where can I find 1160?

3. VAX/Alpha CI

4. Windows installer version

5. SYS$QIOW - Performance Issue

6. How can I use Winsock clients under os/2 TCP/IP ?

7. UCX NFS Client Poor performance issue

8. Flash movioes disappear on iPaq

9. What about performance issues??

10. URGENT: Disk Performance Issue

11. Oracle on VMS performance issue

12. VAX to Alpha Performance/Memory

13. Performance issues - recommendations?