Extremely slow CPU

dreamcompute

#1

Signed up for the beta (thanks for that!) and just spun up a test instance to check out performance, reliability, etc. and I am seeing really quite terrible performance on this spersonic instance. I understand that the Ceph based persistent storage is not the quickest and I would certainly not expect a ton of IO from it, but the sysbench scores are 6-8 times slower than on several other VPS’s both in our own infrastructure as well as other providers with 1/4 the RAM and also on spindle based storage… It really feels like something is wrong here, or is this just on or close to the performance I should be expecting? Any help would be greatly appreciated, thanks!

Please see the below results using sysbench as well as vpsbench (https://github.com/mgutz/vpsbench) on the Dreamcompute instance and a VPS on our infrastructure.

Dreamcompute Supersonic instance on persistent Ceph storage:

sysbench:
root@tester:/home/dhc-user# sysbench --test=cpu --cpu-max-prime=20000 run
sysbench 0.4.12: multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1

Doing CPU performance benchmark

Threads started!
Done.

Maximum prime number checked in CPU test: 20000

Test execution summary:
total time: 157.0727s
total number of events: 10000
total time taken by event execution: 157.0395
per-request statistics:
min: 15.49ms
avg: 15.70ms
max: 102.36ms
approx. 95 percentile: 15.89ms

Threads fairness:
events (avg/stddev): 10000.0000/0.00
execution time (avg/stddev): 157.0395/0.00

vpsbench
2015-05-26 16:03:26 (1.56 MB/s) - written to stdout [4898/4898]

Benching I/O … OK
Benching CPU. Bzipping 25MB file … OK
Benching inbound network. Downloading 100MB file … OK
Share at https://github.com/mgutz/vpsbench/wiki/VPS-Hosts

_05/26/2015

CPU model:  QEMU Virtual CPU version 1.0
Number of cores: 1
CPU frequency:  2199.996 MHz
Total amount of RAM: 2010 MB
Total amount of swap: 0 MB
System uptime:   3 days, 4:04,       
I/O speed:  47.9 MB/s
Bzip 25MB: 38.86s
Download 100MB file: 12.2MB/s

Openvz VPS with 512MB RAM 1CPU (only 1000 units) on standard 7200RPM HDD:

sysbench:
root@monit:~# sysbench --test=cpu --cpu-max-prime=20000 run
sysbench 0.4.12: multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1

Doing CPU performance benchmark

Threads started!
Done.

Maximum prime number checked in CPU test: 20000

Test execution summary:
total time: 23.6586s
total number of events: 10000
total time taken by event execution: 23.6579
per-request statistics:
min: 2.31ms
avg: 2.37ms
max: 7.45ms
approx. 95 percentile: 2.40ms

Threads fairness:
events (avg/stddev): 10000.0000/0.00
execution time (avg/stddev): 23.6579/0.00

vpsbench:
_05/26/2015

CPU model:  Intel(R) Xeon(R) CPU E3-1225 v3 @ 3.20GHz
Number of cores: 1
CPU frequency:  3192.605 MHz
Total amount of RAM: 512 MB
Total amount of swap: 512 MB
System uptime:   5 days, 22:52,
I/O speed:  61.2 MB/s
Bzip 25MB: 3.43s
Download 100MB file: 4.07MB/s

#2

Thanks for posting these informative benchmarks!

I’m not terribly surprised by the CPU speed differences you’re seeing. It looks like the Openvz CPU is running about a third faster than the clock speed of the DreamCompute instance. It’s also running a Xeon core which is going to be a bit faster than our AMD cores emulated through QEMU. We’re working on a new architecture design now and are looking at faster CPUs and ways to increase the performance for everyone.

One thing that might help is to enable swap on the DreamCompute instance. That’s something I noticed right away. We don’t do that by default on our images, but it’s easy to do yourself since you have full root access. https://help.ubuntu.com/community/SwapFaq If you do that, I’d love to see how it affected the results.

That disk I/O is about on par with what I’d expect, though maybe a touch low. Normally I’ve been seeing I/O around 50-60 MB/s and up to 100MB/s in my testing.

It’s also nice to see that the work we’ve done to improve download speeds is paying off. Seeing a 100MB file downloaded at 12.2MB/s is pretty good throughput!


#3

Thanks for the reply!

Still, the CPU numbers on sysbench are extremely slow literally ~8 x slower that the OpenVZ host… Are you sure that is normal?

For fairness here is a KVM VPS w/ 512M 1CPU, it is still over 3 times as faster (with a host CPU of 2.40GHz see below)…

[root@cf1 ~]# sysbench --test=cpu --cpu-max-prime=20000 run
sysbench 0.4.10: multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1

Doing CPU performance benchmark

Threads started!
Done.

Maximum prime number checked in CPU test: 20000

Test execution summary:
total time: 41.6786s
total number of events: 10000
total time taken by event execution: 41.6717
per-request statistics:
min: 4.08ms
avg: 4.17ms
max: 14.41ms
approx. 95 percentile: 4.36ms

Threads fairness:
events (avg/stddev): 10000.0000/0.00
execution time (avg/stddev): 41.6717/0.00

[root@cf1 ~]# cat /proc/cpuinfo
processor : 0
vendor_id : GenuineIntel
cpu family : 6
model : 62
model name : Intel® Xeon® CPU E5-2630L v2 @ 2.40GHz
stepping : 4
cpu MHz : 2399.998
cache size : 15360 KB
fdiv_bug : no
hlt_bug : no
f00f_bug : no
coma_bug : no
fpu : yes
fpu_exception : yes
cpuid level : 13
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss syscall nx pdpe1gb rdtscp lm constant_tsc up pni vmx ssse3 cx16 sse4_1 sse4_2 popcnt lahf_lm
bogomips : 4799.99