MacPro4,1 & MacPro5,1 networking speed seriously limited (Mavericks)

[bxs]

very curious about what engineering has to say about this...

Well so far, I've had three sessions with Apple over this issue. I ran one of their Capture Data programs that gathers up network packet data while I ran some i/o activity from my MP6,1 to/from the MP5,1. Apple's Capture Data program was run on the MP5,1, and the resultant captured data has been sent back to Apple for their software engineers to examine. I do get the impression I have their attention on this issue, and that they want to find the root cause as much as I do.

I now have 3 Macs (MP6,1, MP5,1 and an 27" iMac) in the office and all are connected via double ethernet wires bonded (Virtual Interface via the Network panel configurator). All three communicated with each other through a CISCO SG200 18-port managed switch. All cables are CAT6 and are securely connected to their ports in the switch and in the Macs.

All 3 Macs are running 10.9.3. SMB is enabled on all three Macs.

The only issue is that when reading data from the MP5,1 to the MP6,1 using AJA the read rate is below 10 MBytes/sec and can be as low as 5.3 MBytes/sec during one test.

Reading data from the iMac to the MP6,1 is at wire speed and always tops 100 MBytes/sec and will be as high as 110 MBytes/sec at times.

Writes from the MP6,1 to the MP5,1 and the iMac are always at wire speed and typically are around 107 MBytes/sec.

I will be doing a read/write test later next week with the MP6,1 and MP5,1 directly connected via a single wire. This will eliminate the CISCO SG200 managed switch as being the 'trouble maker'.

The only other thing I can think of is swapping the ethernet cables used for connecting the MP5,1 to the CISCO SG200 switch with brand new ones.

Not sure what else to try and test.... any thought from others ?

BTW... If I disable SMB on the MP6,1 and MP5,1 and then read data from the MP5,1 to the MP6,1 the read transfer rate is improved but never tops 50 MBytes/sec.

 

[mikeboss]

just as I wrote in the first post: only outbound network throughput (from MacPro4,1 and 5,1 perspective) is affected. tell the engineering guys to try with "Intel82574L.kext" V 2.3.0 from OS X 10.8.5 which works flawless and gives approx 100MB/Sec.

BTW: shouldn't you see rates up to approx. 200MB/Sec with bonded ethernet???

EDIT: you definitely should see transfer speeds up to 200+ MB/Sec! I just tested it between my two Mac Pros and speed is 2Gb now. here's the how-to for bonding ethernet: http://support.apple.com/kb/PH14045
and you need to configure aggregated ports on the network switch, too.

 

[bxs]

just as I wrote in the first post: only outbound network throughput (from MacPro4,1 and 5,1 perspective) is affected. tell the engineering guys to try with "Intel82574L.kext" V 2.3.0 from OS X 10.8.5 which works flawless and gives approx 100MB/Sec.

BTW: shouldn't you see rates up to approx. 200MB/Sec with bonded ethernet???

EDIT: you definitely should see transfer speeds up to 200+ MB/Sec! I just tested it between my two Mac Pros and speed is 2Gb now. here's the how-to for bonding ethernet: http://support.apple.com/kb/PH14045
and you need to configure aggregated ports on the network switch, too.

If you are using Activity Monitor those network rates are bogus in regards to telling you the single wire network transfer rates. You need something like XRG for precisely measuring the network data transfer rates... see http://www.gauchosoft.com/Products/XRG/

Bonded ethernet provides failover but not 2x wire speed in one direction. Yes, the throughput can be 200 MBytes/sec if the two wires are used for reads and writes at same time. That is, one wire is transferring and the other wire is receiving. However, both wires cannot be sharing the transfer for reading or writing at same time. The ~200 MBytes/sec for reads or writes is a misunderstood aspect associated with bonded ethernet (Link aggregation or Virtual Interface). Yes, to configure Link Aggregation the ethernet switch must support this as well. My CISCO SG200 18-port switch does support this.

 

[mikeboss]

ok, you're right! I didn't know that. but the software you suggested (XRG) also showed 200MB/Sec transfer speed. in the end, using the good old stopwatch revealed the truth: transferring a large file with or without bonding had absolutely no effect at all, the time needed to finish was exactly the same.

 

[bxs]

ok, you're right! I didn't know that. but the software you suggested (XRG) also showed 200MB/Sec transfer speed. in the end, using the good old stopwatch revealed the truth: transferring a large file with or without bonding had absolutely no effect at all, the time needed to finish was exactly the same.

Just run AJA using a networked attached volume to perform the i/o test. This shows precisely what the read and write rates are across the new work. To use a networked volume you must configure for this is AJA's preferences, otherwise AJA will not recognize the network attached volume/device.

 

[m4v3r1ck]

Just run AJA using a networked attached volume to perform the i/o test. This shows precisely what the read and write rates are across the new work. To use a networked volume you must configure for this is AJA's preferences, otherwise AJA will not recognize the network attached volume/device.

Been using Link Aggregation for over 1 year now: Mac Pro <-> HP 1810-8G <-> Synology DS1812

Does AJA measure a more levelled outcome (see graphs on both)?

Never got the r/w performance any better.

~ Cheers

 

[bxs]

AJA test from MacBook Pro8,3 to MP5,1 with each on 10.9.3

I ran the AJA test from my MacBook Pro8,3 (17-inch, late 2011) to a MP5,1. Each Mac system was running latest 10.9.3, both had SMB enabled and the two Macs were directly connected via ethernet.

With SMB enabled on both systems the read rate was a miserable 5.8 MB/s.

With SMB disabled on both systems the read rate was better but still at half wire speed at 46.7 MB/s.

There is this curious saw tooth shape to the read data transfers as the test ran, and wonder what the significance of that is and what causes it ????

BTW... I ran the exact same test having both Macs connected with ethernet going through an Apple 2TB Time Capsule 802.11ac tower and it made absolutely no difference... the read rate with SMB enabled was below 10 MB/s and with SMB disabled it was up around 45 MB/s.

It seems to me that this network SMB/ON and SMB/OFF read issue is unique to a MP5,1 being the file server and running Mavericks 10.9.3. Has anyone run this test with the read server being something other than a MP4,1 or MP5,1 running Mavericks 10.9.3 ?

 

[bxs]

I note that on my MBP8,3 that the Intel*.kext file is NOT loaded whereas on the MP5,1 it's loaded.

What is the significance of this, and why would this be so ?

Both Macs are running 10.9.3.

 

[mikeboss]

in the windows-world you would call the "Intel82574L.kext" a device-driver. the MacBook has a different NIC built-in, so the kext won't be loaded.

 

[mikeboss]

I just did a test with the system updated to Mavericks 10.9.4 and this issue is still not solved...

 

[oznc]

Solved

I have had the most obscure problem with ethernet speed since the Mavericks upgrade.
Speeds from the MacPro 5,1 to the Xserve suddenly went from 95mb read/110 write to a consistent 54 read/ 54 write sometimes much slower.
I tried the methods in this thread to no avail.
I then tried booting in safe mode and the speeds resumed to the previous 95/110
awesome I thought. Booted back into normal user mode and bam! ethernet not detected!!! This took me down a different road of searching that resulted in the solution to the speed issue.
Boot into safe mode and both ethernet ports are detected no problem. Boot normally and "cable unplugged" is the message in system preferences.
So clearly something is loading in normal boot that isn't in safe mode.

To go back a step we upgraded from 10.8 to 10.9 not a fresh install as there was so much software we didn't want to reinstall on the system.

After much searching a few others had the same problem of ethernet suddenly disappearing and somehow it was tracked down to "caldigit.kext"

Sure enough found it in the /Library/Extensions/CalDigitHDProDrv.kext

Removed it, rebooted and bingo ethernet is back and speeds are back to 95/110.

Now on my MacBook Pro 8,3 the network is perfect even with the CalDigitHDProDrv.kext installed. And this system has never had a fresh install.
Believe it or not I have always (for the last 10 years) when upgrading software or hardware (3rd macbook pro) chosen to restore from previous install all user settings and software.