Re: KVM benchmarks (compared to vmware)

[Joerg Roedel <joro@8bytes.org>]

Does the license of the vmware server you used allow the publication of
your results?

On Mon, Nov 17, 2008 at 01:30:31PM +0100, Martin Vogt wrote:
> 
> 
> Hello list,
> 
> I did some benchmarks over the weekend and compared
> kvm-78 on a 2.6.27.4-2-default against vmware server 2.0.0-122956
> 
> Host Machine:
> - Host E8400 (3GHz) 2GB RAM
> - virtual machines are on iSCSI or NFS
> - virtual machine is booted with -m 1024 and e1000
> 
> 1. Bench
> =========
> 
> Installation of SuSE 11.1: (time to finish)
> 1. iSCSI kvm : 48 min
> 2. NFS   kvm : 54 min (qcow image)
> 3. VMWare    : 38 min
> 
> 2. Bench
> ========
> Booting of SuSE 11.1 beta4 (from grub to getty prompt)
> (with coldcache, http://linux-mm.org/Drop_Caches)
> 
> 1. iSCSI kvm : 33 sec
> 2. NFS   kvm : 44 sec
> 3. VMWare    : 34 sec
> 
> same, with "hotcache".
> 
> 1. iSCSI kvm : 22 sec
> 2. NFS   kvm : 23 sec
> 3. VMWare    : 21 sec
> 
> 3. Bench
> =========
> Now the NIC performance from within the VM to the outer world,
> (measured with tcpspray--program writes to the discard service over tcp)
> 
> VM->outer world
> 
> 1. iSCSI kvm : 12 MB/s
> 2. NFS   kvm : 12 MB/s
> 3. VMWare    : 50 MB/s
> 
> now the reverse:
> 
> outer world->VM
> 
> 1. iSCSI kvm : 50 MB/s
> 2. NFS   kvm : 50 MB/s
> 3. VMWare    : 100 MB/s
> 
> but here the kvm results varies from 30MB/s - 70MB/s
> 
> 
> 4. Bench
> ========
> 
> Now iozone benchmarks, which measure the "virtual drive" of the
> VM: IOZone write a 2GB file to /tmp , writes it again reads it and reads
> it again. (Two runs with 2GB and 4GB file size)
> 
> 1. iSCSI-kvm:
> KB	write	re-writeread	re-read
> 2097152	24131	25349	44325	46055
> 4194304	25056	25419	44917	44654
> avg:	24593	25382	44621	45354
> 
> 
> 2. NFS-kvm:
> KB	write	re-writeread	re-read
> 2097152	7584	37846	43334	42654
> 4194304	12185	20558	37075	40029
> avg     9884	29202	40204	41341
> 
> 2a (because of qcow effect) second run:
> KB	write	re-writeread	re-read
> 2097152	35980	33474	42010	43395
> 4194304	21675	20732	37976	39134
> avg     28827	27103	39993	41264
> 
> 
> 3 VMWARE:
> KB	write	re-writeread	re-read
> 2097152	56168	47829	12568	11688
> 4194304	47315	32088	10897	10801
> avg     51741	39958	11732	11244
> 
> 
> 
> RESULT
> =======
> 
> 
> As a result I would say, that the virtual NIC in VMware is much faster
> (if you see tcpspray as benchmark)
> And the write performace of VMWare is better 50MB/s againt 29MB/s in
> kvm. But VMWare seems to have a horrible read performance (11MB/s vs.
> 44MB/s)
> 
> But in general VMWare is faster, at least during the installation
> benchmarks (38min vs 48 min with kvm)
> Maybe these benches could give an idea for future improvements, but
> I wont say that kvm is slow, its fast enough, but maybe has room for
> improvements.  :-)
> 
> regards,
> 
> Martin
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> --
> To unsubscribe from this list: send the line "unsubscribe kvm" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html