Steve Modica
Forum Replies Created
-
I read the collateral on this and I don’t think it’ll do what you want. Note that in the output for the streams test there’s no mention of latency. So assuming the streams run at a certain speed, he’s going to declare success (or at least, there’s nothing I see in the output that would indicate possible failure).
“Streams” from FCP are different depending on whether they are going over AFP, Samba, Xsan or a local HFS+ filesystem. How you measure them depends on the apps IO size and chosen API for those filesystems.
Steve Modica
CTO, Small Tree Communications -
[Brad Walker] “Thanks for the input but 2 things don’t sit quite right: 1) a 2006 MAC PRO is handling same project perfectly. 2) THis is a 12 core machine…it is SUPPOSED to handle HQ or 4444 or whatever you throw at it. I think 10.6.6 got me in the mess I was in 10.6.4 is working well so far. I’m going to knock on wood.”
My guess would be a problem with the plugins you’re using and optimization. Little code problems that are normally not hit can go bananas on new hardware. The timing is very different because you have more threads, hyperthreading and an integrated memory controller.
The vendor may know of these problems (or perhaps hasn’t tried their code on a 12 core yet!) and you need to let them know. They might be able to do a non-optimized or debug build that works correctly.
Sometimes, things like this are caused by simple mistakes like uninitialized variables in the code. On older machines, they are always 0 anyhow, but on a newer platform that hits memory faster, something might be left in there.
Steve Modica
CTO, Small Tree Communications -
These aren’t system crashes (what I was talking about). This is adobe’s proprietary linker. It’s core dumping. I’d call adobe. They may have a debug version you can load or some thing you can enable that gives them better trace info.
Steve Modica
CTO, Small Tree Communications -
In the old days, we would look at the expanded stack trace and assembly for each call. We would see how the arguments were saved on the stack (usually at a certain offset from the stack pointer), then we could pull them out and see what arguments were being passed up to the next call. In this way, you could find out if a particular structure was corrupted or whatever.
On a mac, we setup two system debugging and have to get into it that way. There’s no crash utility like we used to have. Crash would let us examine the kernel on the fly.
Steve Modica
CTO, Small Tree Communications -
Have you tried swapping graphics cards between systems? Seems like the next test to run.
Also, it was not clear to me whether you actually moved from a 64 bit kernel to a 32 bit kernel, or if you were already 32 bit (I think that’s the default on mac pros). If you hold down the 6 and 4 keys while powering up, it will try and boot 64 bit.
Steve Modica
CTO, Small Tree Communications -
The term “port” is probably being overloaded here. (Meaning it has more than one meaning).
There are standard port numbers in /etc/services and I’m sure they are set to the defaults. People don’t change those.
I’m guessing he app wants ports to be opened in the firewall so traffic can go back and forth. Do you have a firewall setup on the system?
I imagine some ports need forwarding at the firewall so it can get out.
Steve Modica
CTO, Small Tree Communications -
You would need to know what the raid controller was.
Most raids work by writing special headers on the disks. Each card has it’s own format. Even cards with the same chip from different vendors will have different header formats.
When the raid chip sees those headers, it grabs the drives and presents them to the system as a “virtual” device that is the raid. In your case, no chip exists and OS X doesn’t recognize the headers, so it just sees bare drives.
One option would be to try powering the external chassis from another supply (probably just a molex connector. You could get some extensions) or you can go find out what raid controller is in the thing and try to find one you can mount internally (this is extremely unlikely).
Since it’s probably a good time to migrate anyhow, it’s probably a better idea to power the old chassis temporarily, and move everything elsewhere.
Steve Modica
CTO, Small Tree Communications -
You should give us (Small Tree) a call. If you are going to edit from this array using 10bit uncompressed and it’s a feature film (so you will have a complex timeline) you need to be sure the raid operates with low enough latency to sustain the IO in realtime.
I agree with Bob that your life would be a lot easier using Pro Res. If you do this, we could put that same raid on a server and you could have several people editing using the media at once.
Steve
Steve Modica
CTO, Small Tree Communications -
Take a look around for “The Magic Garden Explained”. It’s an old System V unix book, but it covers kernel stack traces and how that works.
Basically, when you have a stack trace, it’s showing you the path of calls that got you to the panic. A panic is something built into the code to protect the system. If the system detects an unexpected condition, it panics to avoid corruption or other problem.
You would generally need to set keepsyms=1 in nvram to get a complete stack trace.
Once you have one, you can generally figure out what module was involved and search around to see if others are seeing the same panic. If the module is part of darwin, you may also be able to look at the source itself.
The most important elements of kernel analysis are understanding kernel internals and having some experience in the type of bugs that occur so you recognize them.
Steve Modica
CTO, Small Tree Communications -
We use things like netpipe and iperf to debug drivers. They aren’t very useful for testing how the network will perform with AFP or Samba.
Apple used to have xsan_tuner which seemed like a serious attempt. I don’t know why they dropped it.
The tools small tree uses are proprietary. They give us a competitive advantage. I can setup 20 or 50 threads on a server and see how it will react. It would be very hard for someone that doesn’t understand kernel IO to setup something like this using final cut.
Steve Modica
CTO, Small Tree Communications