I/O on "ant"
Aug. 18th, 2011 11:24 pm![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
abortrephraseA couple of quick tests:
scp a 3.5GB file from my MBP to ant (the new Atom box, 1000Mb NIC) and red (older Core 2 Duo laptop, 100Mb NIC):
ant: 16.8MB/s, chewing 100% of a CPU core and no iowait
red: 11.2MB/s, chewing ~65% of a CPU core and no iowait
Implication here being that on ant the limiting factor is the single-threaded ssh through which all the data is going, while on red the limiting factor is the NIC.
Investigating ant a little further, rsyncing the same file but over NFS (using unfs3 on ant, OS X Lion on the MBP) it topped out at ~27MB/s for a short while at the start, with CPU peaking at ~50% utilisation, but it didn't take more than about 20 seconds for that to slow way the hell back down to just on 17MB/s, ~30% CPU, and iowaits around the 15% mark.
Which implies that even though one can get rather faster burst out of this box, the internal spindle is just too slow to keep up for very long at all. Which is fair enough, it's a 5400rpm laptop disk.
scp a 3.5GB file from my MBP to ant (the new Atom box, 1000Mb NIC) and red (older Core 2 Duo laptop, 100Mb NIC):
ant: 16.8MB/s, chewing 100% of a CPU core and no iowait
red: 11.2MB/s, chewing ~65% of a CPU core and no iowait
Implication here being that on ant the limiting factor is the single-threaded ssh through which all the data is going, while on red the limiting factor is the NIC.
Investigating ant a little further, rsyncing the same file but over NFS (using unfs3 on ant, OS X Lion on the MBP) it topped out at ~27MB/s for a short while at the start, with CPU peaking at ~50% utilisation, but it didn't take more than about 20 seconds for that to slow way the hell back down to just on 17MB/s, ~30% CPU, and iowaits around the 15% mark.
Which implies that even though one can get rather faster burst out of this box, the internal spindle is just too slow to keep up for very long at all. Which is fair enough, it's a 5400rpm laptop disk.
(no subject)
Date: 2011-08-23 02:22 pm (UTC)I just tested my Zotac (d525), Qnap 412 (arm) and mbp (core2duo), to see how they compared.
Difference is my zotac has a 30GB SSD inside (no spinning rust), and it makes a helluva difference.
The mbp can push 20MB/s to either device, but the qnap pushed 45-65MB/s at the zotac with NFS. Pushing via scp dropped to 6.5MB/s, a side effect of the ARM cpu.
With both transfer methods, the Zotac wasn't loading the cpu threads up much either. both were using around 1 thread @30-40%. I think the d525 might have more guts in it than we think.
I don't have sysstat installed on any of my boxen yet :( will do, once I get dsl back and might do some more profiling.