Sonnet Tempo™ “interrupt storm”
Shootout: Various MacBook Pro Internal Drives from barefeats.com should be of interest to any existing or prospective MacBook Pro user.
Drive speed can be a significant factor in MacBook Pro performance, especially with Photoshop or other programs that open large files. But everything feels perkier when the hard drive is faster, though maxing out the system memory can also help because it allows Mac OS X to scarf up unused memory to use as a large disk cache.
Please see today’s earlier discussion (below) of the Sonnet Tempo “Interrupt Storm”.
For the fifth time today (in only 3 hours or so of usage), the interrupt storm problem kicked in. As usual, I was forced to sleep the PowerMac G5 Quad to make the problem go away. Unfortunately, the Quad never came out of sleep—the fans quickly churned to their highest speed, and my only option was to cut the power and reboot.
Is this too the fault of the Sonnet Tempo card? Perhaps, and perhaps not. But if the Sonnet Tempo card didn’t have the interrupt storm bug, I wouldn’t have been forced to sleep the machine in the middle of my work!
Though not all days are this troublesome, six months of this behavior is enough, and Sonnet’s inability or unwillingness to address the problem in a timely manner is simply unacceptable. My patience has reached its limit. (I was a beta tester, and I have reported the problem many times directly to the engineer responsible for the card firmware, so it’s not like the message has gone unheard).
As readers might know from reading past blog entries [1, 2, 3] and PowerMac G5 Internal Drive Kits, I use two Sonnet Tempo PCI-X cards--an E4i to support four (4) additional internal SATA drives, and an E4P to support four (4) external SATA drives in a FirmTek 2eEN4 external enclosure (PowerMac G5 Quad).
The foregoing setup has performed admirably, except for one irritating problem that Sonnet has not been able to fix: the so-called “interrupt storm”, first reported here on May 15, 2006. Six months of waiting has not resulted in a fix for this problem.
This problem results in nearly an entire CPU core being used to service spurious interrupts from the E4i or E4p. The net result is that on a PowerMac Quad, you lose 20-25% of your processing power, and on a dual-core PowerMac G5, you lose nearly 40-50% of your processing power. Of course, all this activity causes additional heat, which causes the fans to run louder too.
A screen capture from Mac OS X is shown below. Note the 81.4% CPU usage by “kernel_task”, which represents the core of the Mac OS X operating system, within which hardware drivers run:
“Interrupt Storm”
This is what it looks like after the workaround—forcing the computer to sleep, then wake up:
After sleep/wake
I’m disappointed with Sonnet’s attention to this matter, having waited patiently for 6 months for a fix. It happens to me at least once a day, sometimes several. But what if you use your PowerMac Quad for video processing or another application where you can’t stop the middle of a big job (eg video acquisition)? Or just a job where you really need all 4 cores? In a word: bummer.
Apropos: a few hours after I wrote the paragraph above, I started running a CPU and disk intensive job, processing 480GB (yes, 480 gigabytes) of data. For the fourth time today, the interrupt storm problem kicked in, stealing nearly an entire CPU core from my job, and making other activities on the system less responsive.
Performance degradation on a real task
Other PowerMac users report that a recent beta firmware update (available from the Sonnet site) has resolved this issue for them—but it hasn’t resolved it for me. Perhaps it’s because my PowerMac Quad has two of the Sonnet Tempo cards installed. I don’t know, but speaking as a longtime software engineer, I really don’t think Sonnet knows how to fix this problem. Sonnet, will it ever be fixed?
