Comments on: Evaluating new Mac Mini builders, SSDs and -j settings

By: bobthebuilder

bobthebuilder — Mon, 17 Dec 2012 15:43:31 +0000

Maybe drop priority of Mac as platform, and only build once in a while ? most Mac users already believe that Safari on Pentium is 10 times faster than any other browser on IvyBridge….

By: Dan

Dan — Thu, 12 Jan 2012 03:05:38 +0000

“My data also suggests that the formula is (cores – 1) * 4″

As glandium said, you can’t really say that given that you didn’t test with all intermediate values of -j. I agree though that this gives a good enough idea. Once you have chosen the hardware (especially if you go with the quad-core), I’d definitely be interested to see a graph of -j from 1 to 12 (including all values in between) for that machine.

By: IVB

IVB — Wed, 11 Jan 2012 20:07:40 +0000

Mozilla should wait for Ivy Bridge powered Mac Minis. Ivy Bridge uses less power(77w TDP for quad-cores vs Sandy Bridge 95w TDP quad-cores) and between 10-20% more performance.

http://www.anandtech.com/show/4830/intels-ivy-bridge-architecture-exposed

By: John Ford

John Ford — Wed, 11 Jan 2012 19:56:48 +0000

Nope, we don’t use the Mac Minis for building Windows or Linux. We currently have all of Windows and Linux testing on Mac Minis, but we have plans to change that.

By: SMF

SMF — Wed, 11 Jan 2012 18:53:17 +0000

Excellent and thorough analysis jhford

By: glandium

glandium — Wed, 11 Jan 2012 16:16:41 +0000

jhford: Well, as you don’t have data for -j6, you can’t know for sure -j4 is the best for dual core
As for the SSDs, the TS in their reference stands for Toshiba. IIRC they are the bad ones, the least bad ones being SM, which stands for Samsung.

By: John Ford

John Ford — Wed, 11 Jan 2012 16:09:38 +0000

We have 8GB of ram from apple and the SSD is “whatever apple ships”.

From the quad core:
Capacity: 251 GB (251,000,193,024 bytes)
Model: APPLE SSD TS256C
Revision: CJAA0202
From the dual core:
Capacity: 251 GB (251,000,193,024 bytes)
Model: APPLE SSD TS256C
Revision: CJAA0202

Looks like these are the same model. Is this the good model?

Dan, the system shows that there is one processor with four cores on the quad core machine. My understanding is that hyper-threading is on by default in lion and there isn’t a way to turn it off in lion.

Regarding using -j1, -j3, -j5, -j9, I think that’s a great comparison to make. I did some benchmarks with -j1, but they were so much slower than -j2 that they doubled the amount of Y axis i’d need to show. Using cores * 2 + 1 or cores * 1.5 is something we can optimize for once we purchase hardware and start to configure it. The performance degradation seen shows that being off by one -j setting (past -j4) is not a major problem. My data also suggests that the formula is (cores – 1) * 4. I think there is a mix between CPU and I/O which defines how -j settings scale on each machine.

Thanks Armen!

By: Armen Zambrano (@armenzg)

Armen Zambrano (@armenzg) — Wed, 11 Jan 2012 15:08:38 +0000

Excellent job jhford!
You’re going to have a great impact on end 2 end times for developers

By: BarryM

BarryM — Wed, 11 Jan 2012 14:12:10 +0000

Do you use the Mac Minis for other builders like Windows or Linux?

http://www.ocztechnologyforum.com/forum/showthread.php?87950-Enabling-AHCI-for-Windows-on-MBP-2011-now-possible

http://forums.macrumors.com/showthread.php?t=696523

By: glandium

glandium — Wed, 11 Jan 2012 10:06:12 +0000

Dan: I usually use “numer of CPUs * 1.5″