AMD
machine 1
All of these tests were done on an AMD64 X2 4200 w/1G RAM and an IDE hard drive. The host is running Ubuntu Edgy and the guest is a FC5 VM. This machine was running a slightly modified kvm-7 (only to fix the SVM build).
dbench
Using dbench with 10 clients and the default client.txt.
Host: Throughput 291.059 MB/sec 10 procs Guest: Throughput 88.5899 MB/sec 10 procs
scp
scp'ing a large file. Guest is using tun/tap bridged to eth0. scp command is originated from the host, uses no compression, and use the IP address of either the guest or the host.
That is, on the host system, execute either "scp largefile hostip:" or "scp largefile guestip:"
Host: Sustained transfer rate of 15.5 MB/sec Guest: Sustained transfer rate of 2.9 MB/sec
PIO latency
Measuring the number of cycles before and after an outb 0x81. 100 delta's were used:
Mean: 5657 Standard Deviation: 556
machine 2
Is an IBM BladeCenter LS21/LS41 -[797252G] with 2 Dual-Core 2.4 Ghz AMD Opteron 8216 HE, 4 GB of ram and sas disks.
The host system is a debian sarge amd64 running with an home-made 2.6.18.3 kernel and kvm 6
Kernel compile time
Unfortunately for this test, the guest system and the host system diverges on several points:
software: the guest system was a debian sarge running with a default debian kernel 2.6.17
hardware: the guest system had only 512 MB of RAM
Kernel compile time for a kernel 2.6.19 using make-kpkg (same gcc, same libs, same config):
Host system: 31m32s Guest system: 84m1s
machine 3
Is an AMD64 X2 3800+ w/2G RAM and an IDE hard drive. The host is running Gentoo and the guest is a Knoppix 5 VM.
nbench
VANILLA NBENCH ON KNOPPIX 5.0 - KVM (svn 12/20/2006 2am)
CC: gcc 4.0.4 20060507 (prerelease) (Debian 4.0.3)
VANILLA NBENCH ON KNOPPIX 5.0 - CURRENT QEMU (0.8.2) + kqemu
CC: gcc 4.0.4 20060507 (prerelease) (Debian 4.0.3)
VANILLA NBENCH ON KNOPPIX 5.0 (NATIVE)
CC: gcc 4.0.4 20060507 (prerelease) (Debian 4.0.3)
|
VS |
P90 |
kqemu |
kvm |
|
INT |
56.220 |
(+7.6%) |
(+8.8%) |
|
FP |
40.717 |
(+6.8%) |
(+8.2%) |
|
K6/233 Index |
|||
|
MEM |
15.381 |
||
|
INT |
13.094 |
||
|
FP |
22.583 |
||
INTEL
machine 1
Is an Apple macbookpro with a 32bits dual-core intel T2500 @ 2.00GHz with 1GB RAM and a 100 GB sata hard-drive, running ubuntu edgy with a 2.6.19.1 and kvm-7
Guest is a debian sarge running with a 2.4.27 and 512 MB RAM
dbench
Using dbench 3.04 with 10 clients and the default client.txt.
Host: Throughput 393.178 MB/sec 10 procs Guest: Throughput 157.592 MB/sec 10 procs
machine 2
Is an apple macbook with a 64bits intel core 2 duo and 1GB RAM.
Below it the compilation time of a linux 2.6.19 kernel allno config
| real time | user time | system time
qemu | 46m19s | 21m19s | 1m41s
kvm | 4m51s | 2m9s | 0m16s
kvm-mmu | 9m17s | 3m40s | 0m57s
Machine 3
Disk subsystem: Infortrend S16S-G1030
OS: Debian Lenny 2.6.25-2-amd64, KVM-70
Guest Lenny: (Unable to use virtio for disk since it keept freezing the guest)
Using dbench 3.04 and 10 procs, running default clients.txt
Host: Throughput 1648.3 MB/sec 10 procs Guest Sid: Throughput 312.144 MB/sec 10 procs Guest Lenny: Throughput 310.835 MB/sec 10 procs
Using tbench 3.04 and 10 procs, running default clients.txt
Host Debian lenny: Throughput 80.5566 MB/sec 10 procs Guest lenny 2.6.25-2 virtio: Throughput 43.1615 MB/sec 10 procs Guest sid 2.6.26-1 virtio: Throughput 56.6345 MB/sec 10 procs
I used iozone default -a benchmark, I lowered the ram on the guest to 384MB to make sure it could not keep the entire dataset in ram.
Full data in openoffice-format.
