QUESTION Benchmarking ARM Processors

[tycho]

I finally got around to tweaking my old Dockstar so that it handles the I-PBX image (see http://nerdvittles.com/?p=10560) more gracefully. (fn1) I very much like my cheapie Pogo Mobile -- best $10 PBX I ever had -- and find that it is uber reliable and essentially bullet-proof. But, it has a processor clock speed of 800MHz, whereas my much-older Dockstar using the same architecture has a clock speed of 1.2GHz -- 50% faster.

Given that the GUI can be pretty sluggish on the Pogo (but still certainly usable if you don't change things often), I nevertheless thought it might be cool to replicate my Pogo set-up on the Dockstar to see just how much faster it might be. (fn2) This went swimmingly using the incrediblebackup and incrediblerestore scripts: I was able to clone the Pogo to the Dockstar with no issue.

I suppose I could use a stopwatch to time things like an asterisk reload or a trunk revision. But I'd rather see if anyone is aware of any benchmarking programs/software that I could use on these ARM5 machines to see (just for fun) if the Dockstar's 50% clock-speed advantage translates into anything remotely like that in tests.

______________________
(fn1: My Dockstar, first put online in early 2011, has had a series of 'U-Boot' installations. Even though it has a Kirkwood ARM5 proc. like the Pogo, I'd had a number of problems booting the I-PBX Image. It turns out that my older U-Boot did not nicely support the 3.14.X kernel of the Pogo image. So, I finally updated the Dockstar to a U-Boot from 7-2014 which made the problem go away.)

(fn2: I used the Dockstar way-back-when, first compiling asterisk on the device and adding native FreePBX, based on advice from user twinclouds and others found in myriad forums -- mostly the Jeff Doozan forum. I moved away from it because I was changing things constantly, prompting me to move to a bare-metal cast-off PC and PIAF, then VMs, then the cloud, if for no other reason than the speed advantage. I'm older now, and tinker less, so going back to slower is fine. )

 

[tycho]

Dug around and answered my own question. I was successfully able to run the Crowd9 benchmark script that posts to Serverbear on the Pogo-ized Dockstar. After about 18 hours (don't know the exact start time; didn't pay attention) I got the results. Just for grins I ignored sound advice and ran the benchmark concurrently with everything else that was running (Asterisk, IPBX, and all the associated programs and daemons). The verdict:

Distro:
Debian GNU/Linux 7 \n \l
CPU Info:
processor    : 0
model name    : Feroceon 88FR131 rev 1 (v5l)
Features    : swp half thumb fastmult edsp
CPU implementer    : 0x56
CPU architecture: 5TE
CPU variant    : 0x2
CPU part    : 0x131
CPU revision    : 1
 
Hardware    : Seagate FreeAgent DockStar
Revision    : 0000
Serial        : 0000000000000000
Disk space:
Filesystem    1K-blocks    Used Available Use% Mounted on
rootfs          3810728 2049892  1567256  57% /
udev              10240      0    10240  0% /dev
tmpfs              12268    228    12040  2% /run
/dev/sda1        3810728 2049892  1567256  57% /
tmpfs              5120      0      5120  0% /run/lock
tmpfs              24520      0    24520  0% /run/shm
tmpfs              61332    636    60696  2% /tmp
total            7734936 4100648  3247128  56%
Free:
            total      used      free    shared    buffers    cached
Mem:        122664      94856      27808          0      1696      34568
-/+ buffers/cache:      58592      64072
Swap:      262140      16104    246036
 
------------------------------------------------------------------------
Benchmark Run: Wed May 20 2015 07:50:07 - 08:18:43
1 CPU in system; running 1 parallel copy of tests
 
Dhrystone 2 using register variables        2004541.1 lps  (10.0 s, 7 samples)
Double-Precision Whetstone                      44.3 MWIPS (9.9 s, 7 samples)
Execl Throughput                                488.5 lps  (29.8 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks        81096.5 KBps  (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks          27275.7 KBps  (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks        159421.5 KBps  (30.0 s, 2 samples)
Pipe Throughput                              168408.7 lps  (10.0 s, 7 samples)
Pipe-based Context Switching                  15723.0 lps  (10.0 s, 7 samples)
Process Creation                              1401.6 lps  (30.0 s, 2 samples)
Shell Scripts (1 concurrent)                    825.7 lpm  (60.1 s, 2 samples)
Shell Scripts (8 concurrent)                    102.1 lpm  (60.5 s, 2 samples)
System Call Overhead                        564591.2 lps  (10.0 s, 7 samples)
 
System Benchmarks Index Values              BASELINE      RESULT    INDEX
Dhrystone 2 using register variables        116700.0    2004541.1    171.8
Double-Precision Whetstone                      55.0        44.3      8.1
Execl Throughput                                43.0        488.5    113.6
File Copy 1024 bufsize 2000 maxblocks          3960.0      81096.5    204.8
File Copy 256 bufsize 500 maxblocks            1655.0      27275.7    164.8
File Copy 4096 bufsize 8000 maxblocks          5800.0    159421.5    274.9
Pipe Throughput                              12440.0    168408.7    135.4
Pipe-based Context Switching                  4000.0      15723.0    39.3
Process Creation                                126.0      1401.6    111.2
Shell Scripts (1 concurrent)                    42.4        825.7    194.7
Shell Scripts (8 concurrent)                      6.0        102.1    170.2
System Call Overhead                          15000.0    564591.2    376.4
                                                                  ========
System Benchmarks Index Score                                        121.9
------------------------------------------------------------------------

For grins, here is a comparison run of my bottom-dollar Cloud at Cost IPBX 12 instance, obtained while running all of the IPBX-related programs and daemons, as well as rTorrent:

Distro:
Ubuntu 14.04.1 LTS \n \l
CPU Info:
processor    : 0
vendor_id    : GenuineIntel
cpu family    : 6
model        : 26
model name    : Intel(R) Xeon(R) CPU          L5520  @ 2.27GHz
stepping    : 5
microcode    : 0x16
cpu MHz        : 2266.747
cache size    : 8192 KB
physical id    : 0
siblings    : 1
core id        : 0
cpu cores    : 1
apicid        : 0
initial apicid    : 0
fpu        : yes
fpu_exception    : yes
cpuid level    : 11
wp        : yes
***
bogomips    : 4533.49
clflush size    : 64
cache_alignment    : 64
address sizes    : 40 bits physical, 48 bits virtual
power management:
Disk space:
Filesystem                1K-blocks    Used Available Use% Mounted on
/dev/mapper/CAC_VG-CAC_LV  9847336 4556500  4778468  49% /
none                              4      0        4  0% /sys/fs/cgroup
udev                        239652      4    239648  1% /dev
tmpfs                        50180    544    49636  2% /run
none                          5120      0      5120  0% /run/lock
none                        250896      0    250896  0% /run/shm
none                        102400      0    102400  0% /run/user
/dev/sda1                    136709  36397    90134  29% /boot
total                      10632297 4593445  5516306  46% -
Free:
            total      used      free    shared    buffers    cached
Mem:        501796    443984      57812        36      14124    275612
-/+ buffers/cache:    154248    347548
Swap:      976892    271116    705776
 
------------------------------------------------------------------------
Benchmark Run: Tue May 19 2015 17:54:32 - 18:23:10
1 CPU in system; running 1 parallel copy of tests
 
Dhrystone 2 using register variables      22545485.7 lps  (10.0 s, 7 samples)
Double-Precision Whetstone                    2891.9 MWIPS (9.8 s, 7 samples)
Execl Throughput                              2126.0 lps  (29.9 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks        336194.6 KBps  (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks          96556.3 KBps  (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks        879109.9 KBps  (30.2 s, 2 samples)
Pipe Throughput                              562302.1 lps  (10.0 s, 7 samples)
Pipe-based Context Switching                137562.2 lps  (10.0 s, 7 samples)
Process Creation                              6001.9 lps  (30.0 s, 2 samples)
Shell Scripts (1 concurrent)                  4243.2 lpm  (60.1 s, 2 samples)
Shell Scripts (8 concurrent)                    546.7 lpm  (60.1 s, 2 samples)
System Call Overhead                        432903.4 lps  (10.0 s, 7 samples)
 
System Benchmarks Index Values              BASELINE      RESULT    INDEX
Dhrystone 2 using register variables        116700.0  22545485.7  1931.9
Double-Precision Whetstone                      55.0      2891.9    525.8
Execl Throughput                                43.0      2126.0    494.4
File Copy 1024 bufsize 2000 maxblocks          3960.0    336194.6    849.0
File Copy 256 bufsize 500 maxblocks            1655.0      96556.3    583.4
File Copy 4096 bufsize 8000 maxblocks          5800.0    879109.9  1515.7
Pipe Throughput                              12440.0    562302.1    452.0
Pipe-based Context Switching                  4000.0    137562.2    343.9
Process Creation                                126.0      6001.9    476.3
Shell Scripts (1 concurrent)                    42.4      4243.2  1000.8
Shell Scripts (8 concurrent)                      6.0        546.7    911.1
System Call Overhead                          15000.0    432903.4    288.6
                                                                  ========
System Benchmarks Index Score                                        664.0
 
------------------------------------------------------------------------

I guess I'm thinking that an overall score of 121.9 for a 5-year-old embedded 1.2GHz ARM processor doodad with a 4GB giveaway-from-Microcenter-no-name USB stick compares reasonably favorably to an overall score of 664.0 for the cloud-instance PBX with the Intel(R) Xeon(R) CPU running at 2.27GHz with 512MB RAM and a 10GB SSD disk...

Next step: run the benchmark on the PogoMobile so as to be able to compare it to the Dockstar; my initial goal.