Measuring client speed

[twickland]

Does anyone know what the difference is between OS X and Windows clients as to how Retrospect measures client speeds?

 

In our network configuration, all our Windows clients (NT, 2000, XP) reveal a numerical client speed in the Get Info window.

 

None of our OS X clients (10.2, 10.3, 10.4; client versions 6.0.108 - 6.1.107) show anything other than zero for client speed.

 

I'm pretty sure that whatever is causing our inability to measure client speed of the OS X clients is also what is causing us to get a 515 Piton protocol error when Retrospect attempts to resume backing up an OS X client after changing to a new backup set member.

 

If I know more about what the issue is, I may be able to talk our network guru into adjusting his security parameters to enable the communication necessary to avoid these problems.

[waltr]

Hi Twick,

 

I'm not sure what could be affecting your Macs in this way but I think you are right to think it is related to your 515 errors.

 

On my G5 I can see both Windows and Mac clients, and doing a 'Get Info' gives me speeds from both OS's.

 

If you want to eliminate the network you could try connecting a Mac Client to the server via a crossover cable and check to see if you can see the speeds. Maybe that would help your network guy to see that there is a problem.

[twickland]

Waltr:

 

Thanks, but I'm afraid you misunderstand our situation. I _know_ that the network configuration is the source of my problem, but my goal and the goal of the network manager are somewhat opposed. For security reasons, he wants to restrict as much as possible the kinds of communication that can pass through the switches, while still allowing traffic necessary for people to be able to do what they need (though not necessarily in the most convenient way). From his perspective, the specific problems I'm seeing are minor, and I frankly would have to agree with him.

 

If, however, I can pinpoint more clearly what communication needs to be able to pass, he may be able to configure the network in a way that will allow me to avoid these particular errors without unduly compromising the security he wants to impose.

 

Thus my question: what is different in the way that Retrospect communicates with OS X clients vs. Windows clients as to how it measures client speed?

 

Incidentally, though we no longer have any OS 9 clients, when we did have them, Retrospect was also able to measure their speeds.

[twickland]

Let me try once again to rouse the experts at EMC^2 Dantz:

 

By what mechanism does Retrospect determine client communication speed for OS X clients?

 

By what mechanism does Retrospect determine client communication speed for Windows clients?

 

(According to our network guru, UDP packets generated in-house are passed unrestricted across our switches, whereas ICMP and subnet broacast are restricted.)

[natew]

Hi

 

Multicast and subnet broadcast are used for client discovery. After that it is a standard TCP connection. I'm pretty sure the speed test is just a standard ping for Windows clients.

 

I don't see a speed measurement when I log in Mac clients. I don't think that feature exists. The 0 number is the clock offset, not speed.

 

Thanks

nate

[CallMeDave]

Quote:

---

I don't see a speed measurement when I log in Mac clients.

 

---

 

I do.

 

Configure->Clients->YourClient->File Menu->Get Info

 

Information includes:

 

Memory

System

Echo time

Speed

Clock offset

 

The client I'm looking at has a value of "Speed: 2400 KB/s"

 

Dave

[natew]

 

Ah yes, I see now.

Thanks Dave!

 

Twickland,

Is this the screen you were looking at? It works for me too.

 

Thanks

nate

[twickland]

Yes, the Get Info window is what I look at.

 

With either client, Echo time is always 0.0 (not sure whether I fully believe this).

 

With Windows clients, Speed is typically between 4,400 and 11,000 KB/s.

 

With OS X clients, Speed is always indicated as 0 KB/s (which, of course, is nonsense).

 

Clearly, Retrospect uses a different protocol for measuring client speed for Windows and OS X clients. Knowing what this difference is might help us be able to enable whatever communication is necessary to properly measure speed of OS X clients.

 

Not being able to measure client speed prevents us from using a speed threshold in backups, which is a minor annoyance.

 

More critically, it is also implicated in causing backups of OS X clients to fail with a 515 Piton Protocol error whenever the backup pauses to begin writing to a new tape member. In addition to making the client wait another night for backup, this also means that the (typically) many GB of files just backed up to the earlier tape member never get compared.

 

(Interestingly, pausing the backup manually does not seem to generate any errors; it's only the pause at tape member change that has this effect.)