tcpdump mailing list archives
Re: Libpcap performance under VMWare guest OSes
From: Rick Jones <rick.jones2 () hp com>
Date: Thu, 10 Dec 2009 16:50:23 -0800
Mark Bednarczyk wrote:
I hope you don't mind if I submit ttcp numbers instead of netperf.
in and of itself, no, but ttcp doesn't give pps figures, it gives bulk transfer figures - which is one of the reasons netperf was created in the first place :)
the next "wonder" is how much CPU is left over when the guest(s) are running at these rates.
rick jones
The following tests are between 3 physical machines on a 10-meg ethernet network all connected to the same subnet. System #1: win_32 and ubuntu1_32 (dual CPU) System #2: fedora core 64-bit (quad CPU) System #3: debian_32 and ubuntu2_32 (dual CPU) (I also have various other OSes solaris, freebsd, etc but I think this mix is more then enough for our purposes.) For file copies via sftp I get: host(win_32) to guest(ubuntu1_32) (same physical machine and disk): 20.8Mbps host(win_32) to host(fc_64) (different machines over 10 meg eth): 6.8Mbps host(win_32) to guest(debian_32) (different machines over ethernet): 6.5Mbps ttcp transfers: host(win_32) to host(fc_64) (different phsyical machine): 8Mbps host(fc_64) to guest(ubuntu2_32) (different machine): 7.8Mbps host(fc_64) to guest(debian_32) (different machine): 7.7Mbps guest(debian_32) to guest(ubuntu2_32) (same physical machine): 624Mpbs So it looks like that the network drivers on VMs are able to handle quiet a bit of traffic upto 600Mbps when a physical network is not involved, but respectible 7 - 8 Mbps on a 10-meg ethernet network. Somehow libpcap, when it taps into this captured traffic, is not able to handle a fraction of the actual traffic. Cheers, mark...-----Original Message----- From: tcpdump-workers-owner () lists tcpdump org [mailto:tcpdump-workers-owner () lists tcpdump org] On Behalf Of Rick Jones Sent: Thursday, December 10, 2009 4:43 PM To: tcpdump-workers () lists tcpdump org Subject: Re: [tcpdump-workers] Libpcap performance under VMWare guest OSes What is the delta in "plain" packet per second performance between the VMguest and bare iron? I'd expect that to correlate with libpcap performance. Ie, if the VMguest cannot do well on plain packet per second stuff (like say burst mode/aggregate netperf TCP_RR) then it won't do well on libpcap. rick jones - This is the tcpdump-workers list. Visit https://cod.sandelman.ca/ to unsubscribe.- This is the tcpdump-workers list. Visit https://cod.sandelman.ca/ to unsubscribe.
- This is the tcpdump-workers list. Visit https://cod.sandelman.ca/ to unsubscribe.
Current thread:
- Libpcap performance under VMWare guest OSes Mark Bednarczyk (Dec 10)
- Re: Libpcap performance under VMWare guest OSes Rick Jones (Dec 10)
- Re: Libpcap performance under VMWare guest OSes Mark Bednarczyk (Dec 10)
- Re: Libpcap performance under VMWare guest OSes Rick Jones (Dec 10)
- Re: Libpcap performance under VMWare guest OSes Guy Harris (Dec 10)
- Re: Libpcap performance under VMWare guest OSes Mark Bednarczyk (Dec 10)
- Re: Libpcap performance under VMWare guest OSes Guy Harris (Dec 10)
- Re: Libpcap performance under VMWare guest OSes Mark Bednarczyk (Dec 10)
- Re: Libpcap performance under VMWare guest OSes Mark Bednarczyk (Dec 10)
- Re: Libpcap performance under VMWare guest OSes Rick Jones (Dec 10)
- Re: Libpcap performance under VMWare guest OSes Mark Bednarczyk (Dec 10)