Chris Murphy
Forum Replies Created
-
I don’t understand the proposed topology. The Sonnet adapter is connected to computers. Assuming you want 10GigE for all computers, you’ll need three of them, one for each computer, with the mini connected to the Sonnet adapter which is in turn connected to the Pegasus.
Be aware that to support LACP 802.3ad to bond those two 10GigE ports into a single 20GigE logical link is often non-trivial. You’ll need a 10GigE switch that explicitly supports this and is compatible with how OS X implements it.
-
I said from the outset that without hardware (a NIC) there’s no hardware offloading. The work the NIC hardware would otherwise do is being done in the kernel. If you disagree with some aspect of that, address that separately.
Twice you’ve referred to TCIP protocol, so now I’ll have to assume it’s not a typo and I’ll ask you to define that acronym because I don’t understand it. If you mean TCP/IP (they are two different things, they’re not the same protocol) then what you’ve written doesn’t make much sense. Airport cards typically don’t offload TCP/IP checksumming, this gets done in software, and yet packet loss results in retries (when using TCP, not UDP of course), so this wouldn’t make any difference with TCP/IP over Thunderbolt (minus NIC).
Also I advise not passively calling someone a moron by stating the only two possible reasons for miscommunication is they’re not reading what you’ve written or that they’re the one missing something. It has a tendency to degrade the conversation.
-
Why would there be packet loss with a properly built network? And how does checksum offloading mitigate packet loss? My answers are, it wouldn’t, and it doesn’t. The issue is strictly performance because at 10GigE speeds (possibly 20Gbps since it’s full duplex) it’s a massive amount of checksumming occurring. This siphons CPU cycles away from all video editing tasks when done in software than hardware since it’s the kernel that has to do that checksumming work, instead of the NIC.
-
The I/O scheduler is a function of the kernel. For a VM arrangement that includes the guest and host VMs. It gets really tricky optimizing this because the caching policy for the VM makes a difference which I/O scheduler has the biggest effect. And it also matters what file system is being used. If you’re using XFS, deadline scheduler is commonly recommended on the XFS mailing list. So it’s easiest to just use the boot parameter elevator=deadline on both the host and guest.
-
The bigger confusion is USB 3.1 Gen 1 vs USB 3.1 Gen 2. Apple’s USB-C uses USB 3.1 Gen 1, which is 5Gbps bandwidth, not the 10Gbps bandwidth offered in Gen 2.
-
Any possible valid criticism or argument you have is totally obliterated by comment #7 and makes you look like a violent nut case. People who assault other people are properly put in jail, shortly followed by therapy, and children are people too.
-
Since there’s no hardware, and thus no hardware offloading, it’s all being done in the kernel. So Thunderbolt 1 vs 2 isn’t the issue. Clearly the hardware has the ability to do the work when using 10GigE physical cards.
-
Weird.
Anyway, this sort of thing requires RAID logs, and often kernel messages are helpful also. These days kernel messages get dumped into the system.log and can be filtered by typing the word kernel in the search field. Then what shows up can be copy/pasted into a separate text file and then put up on pastebin. Otherwise it’s like entering a pitch black room and someone says “there’s a problem, what is it?”
-
Disconnects are a different problem, unrelated to stripe size. You need to post the logs somewhere, and/or get the logs to support and have them take a look. It could be as simple a fix as a firmware update, or it might be a hardware defect requiring some hardware to be swapped out. But a disk icon vanishing means it’s not being properly unmounted, and sounds like the physical device is being yanked out from under the OS, and as such your data is at risk of being corrupted. I’d get it dealt with sooner than later.
-
4KB alignment is a concern with outdated software. Mainly it’s Windows XP era software, and some Linux utilities before 2007 that could have an issue. But it’s long since patched for 1MB alignments on both Windows and Linux.
On PowerPC computers, it defaults to use Apple Partition Map which I think can handle a 6TB drive like GPT can; but I’m not sure what it uses for alignment off hand.
The bigger issue is whether these are SMR drives, I don’t expect them to be suitable for video production at all if they are. Possibly useful as nearline storage instead of or supplementary to tape. Think of it as being like tape for writing with much faster seek times; and equivalent to the random access of a regular hard drive for reads.