UCX Memory

[Rajarshi Gupta]

Hi ,

We are using Open VMS 7.1 with UCX V4.2 ECO 5 on one of our VAX 4000 server. We are experiencing some high non zero wait value
( UCX show Comm / Mem ) for large buffer but it is not increasing cumilatively like error counts. Neither it is changing dynamically. For example it is staying at say 4502 for One and half hour then suddenly jumping to 4732 and staying there for some hour again coming back to 589 range for two hours... like this.
could you please explain how I can use these values to quantify the error? Is this value representing the wait for that particullar instance? Does it get reset at Midnight with other VMS accounting information? Much you could provide information on this topic will be more helpful for me.

Thanking on anticipation

Raj

[Wim Van den Wyngaert]

This old document might help.
Search for "wait".

Wim

[Antoniov.]

Hi Raj,
here there is TCP/IP troubleshoting guide of V5.3; I hope it can help you.

Antonio Vigliotti

[Wim Van den Wyngaert]

Antonio : where ?

Note that 5.3 memory management is completely different.

Wim

[Volker Halle · Accepted solution]

Raj,

the Waits counter value should not 'decrease', except if it overflows.

Compare the Peak buffer values with the permanently allocated ones and consider to increase the permanent values, if the peak values are near or reach the max values. As far as I remember, the maximum values you define, need to be a multiple of the appropriate MIN= values.

Volker.

[comarow]

There is a UCX tuning guide, which is obsolete in TCPIP. Log a call and we can send it to you.

[Rajarshi Gupta]

If we increase " QUOTA " for send and receive for UCX will it help?

I have checked the quota with the command -UCX show prot tcp/ param

and it is showing the default value for both send and receive.

Is this " quota " related with UCX memory for out-bound and in-bound connections? Do I need to increase the quota before increasing the Max value for " Large Buffer " & " small Buffer " ?

[Volker Halle]

Rajarshi,

please look at the 'Tuning and Troubleshooting Hints to Improve UCX Performance' document posted by Wim.

If you are seeing Large Buffer and Small Buffer waits, you need to increase the number of these types of buffers (MBUFs) as provided by the UCX kernel:

UCX> SET CONF COMM/SMALL=(MIN:n, MAX:m)

same with LARGE (and/or IRP).

You will need to shutdown and restart UCX and you may also need to tune your system using AUTOGEN, as these buffers consume nonpaged pool. This is all described in the above document.

If you are unsure what values to use, please provide the output of UCX SHOW COMM, UCX SHOW COMM/MEM and SHOW MEM/POOL/FULL and we'll try to come up with the exact commands for you to use.

Increasing the QUOTA values won't help at all.

Volker.

[Rajarshi Gupta]

Dear Volker,

Thanks for your reply. Could you please let me know what is the use of the QUOTA values ? Any elaborate explanation on QUOTA will be highly appreciated.

Thanks
Raj

[Volker Halle]

Rajarshi,

the use of QUOTAs in OpenVMS generally limits the ability of a single instance (process, user, job, session etc.) to deplete a system-wide resource.

In this case in UCX, it limits the TCP message queue size for Receive/Send messages.

If you are seeing UCX MBUF buffer waits, this mean that the UCX kernel needs a packet from a memory buffer and none is available, so it has to wait. This will cause a little bit of delay in your TCPIP communication. If you don't want this to happen and have enough nonpaged pool, you can increase the number of MBUFs available to the UCX kernel.

Increasing the QUOTAs without increasing the underlying buffer space, may even make things worse.

Volker.

[Wim Van den Wyngaert]

Raj,

Found this :
Before actual transmission, a transmit data operation moves data from the user process space to the large buffer.
Until a user process is ready to read data, a receive data operation stores received data in the large buffers.

Is it possible that user processes doing network traffic are on low priority and thus do not get the cpu to receive the packets ?
Is it possible that a cpu eater is active from time to time ?

Wim

[Rajarshi Gupta]

The default value for the quota in our system is 4096 and max value for small & large buffer is 1000 and 200. are these all denotes the value in Blocks ? If QUOTA is the limit for individual process / user or instance how can it be 4096, greater than Max large buffer which is 200?

[Volker Halle]

Rajarshi,

TCP receive and Send QUOTA is measured in bytes.

The data shown with UCX SHOW COMM [/MEM] is the no. of buffers (small, large, IRPs).

Small mbufs are 256. bytes, large mbufs vary in size (see the document provided by Wim).

Volker.