openvms max throughput for DEGXA nic

[Layne Burleson_1]

Does anyone know the maximum or average throughput that you can expect from an Alpha server DS10 running a DEGXA NIC at 1000 full duplex for OpenVMS 7.3-2?

[Hoff]

I'm not at all sure there's any way to truly answer this other than in theoretical terms. eg: GbE speeds.

If pressed for numbers, I'd tend to put a Broadcom 5700-class (Broadcom 5703 (DEGXA) or the Broadcom 5701 (A6825A)) NIC at somewhere around 120 megabytes per second (sustained) in the right box with the right PCI slot with the right application, depending on a whole host of factors (pun fully intended) including the packet sizes and all the various potential sources of contention. And there are any number of things that can preclude reaching that speed.

There are factors all the way up the I/O stack (eg: the selected network transport) and all the way out the LAN (eg: which switch and which router). Not to mention the available bandwidth within the application itself, as folks very seldom use a GbE NIC in isolation.

Put simply, use the DECnet DTS/DTR tools or other load-testing tool -- and try it in your own configuration. See if you get anywhere near the number above, or near the theoretical bandwidth numbers...

Stephen Hoffman
HoffmanLabs LLC

[Jon Pinkley]

That's an "it depends" type question.

Are you wondering what can be done in a lab with specially written programs, or were you wondering what to expect for MSCP served devices or FTP transfters, etc.

Richard will have probably have some of the "maximum" type answers, but there are too many variables to be able to give a good answer that will apply to your situation.

Are you thinking of replacing another NIC to get better performance? What type of application are you going to use it for?

How is the DS10 connected to the other devices it will be communicating with? Does the path support jumbo frames, and is the traffic able to make use of the larger MTU possible with Jumbo frames.

What is going to be feeding the cards? The point being that if you are doing an image backup of a fragmented disk with many small files to a saveset on other box, the disk may be your bottleneck.

Also, if you have small packets, you may be able to get a very high number of packets per second through the NIC, but you may not have much CPU left for real work.

Can you answer the same question for the NIC you currently have in the DS10?

Jon

[Layne Burleson_1]

I realize that I submitted a very generic question that is incredibly had to answer. We have just installed the gb Ethernet card replacing a 10/100. No, we have not enabled "jumbo frames" as I'm not exactly sure how to do that or what that involves. We are just having some continued performance problems with Veritas Netbackup and FTP transfers. No real improvement since installation of NIC. Network folks seem to think everything is fine and configured properly on their end. We got sniffers running to determine our actual speeds. Is there anything I need to update on the OS side after installing the card?

[Hein van den Heuvel]

I've seen 80MB/sec between my DS10 and workstation from cache to a null device.

Tried again just now, but in the mean time I installed an Intel based AD331A (Gigabit Ethernet). This time I was reading 30 MB/sec, from VMS XFC cache to local Sata disk. This was with the UCX TCP stack, regular FTP, Jumbo Frames, patched OpenVMS 8.3 and a 200 MB input file set to fixed length 1024 recs.

That caused 20% Interupt, 9% Kernel, 8%Exec (RMS) and, 1% User mode. Lame!

The run before I was 100% CPU bound, had roughly 0MB/sec, with 88% exec mode, 11% usermode and 1% interupt statck.

Exactly the same data bytes, all zero, from the same cache. Why? The file was set to variable length records and even though I spcified 'bin' to FTP, it was doing record IO !? reading 100 million times 0 bytes at a time, moving 2 bytes forward each get.

Broken!

But mostly it goes to show that 'it depends'.
What is the source MB/sec delivering power?
How good is the suction power power for the destination? What are the products & versions involved in the stack.

Regardless... with no specific tunig you should expect to well exceed the 100mb max of 12MB/sec.

btw... my FTP session was using RMS settings MBC=50., MBF=10 ???
Those are odd settings, and for now I do not kwow where they came from. My RMS defaults are MBC=120.,MBF=4. Does anyone happen to know saving me the excursion?

Hope this helps some,
Hein van den Heuvel (at gmail dot com)
HvdH Performance Consulting

[Robert Gezelter]

Burleson,

As Hoff, Hein, and Jon have observed, there are a wide range of variables. Certainly, at gigabit speeds, jumbo frames will help substantially.

To be definitive, it would be necessary to look both at system performance data and network utilization data. In my experience, "network throughput" can have any number of causes, from the truly mundane to the exotic.

- Bob Gezelter, http://www.rlgsc.com

[Layne Burleson_1]

I'm going to move this discussion over to another thread that I have open: https://forums11.itrc.hp.com/service/forums/questionanswer.do?threadId=1240404