xen that I was blocked on

HVM options
nic deprecated
vif rate options

checking help_config

Tests running in xen express

baseline with 0.3% degradation
fork with misbehaved one 59.93%good 60.29%tolerant. Specweb survives and ctrl+c can bring the system back.
out of memory: killed process 11740(apache2)
Next is memory, the same as fork, the specweb survives and ctrl+c can bring the system back.
malloc.sh line7 killed ./malloc.sh -i
out of memory: killed process 3090(mallocBomb) score 15538 and children.
Then is DISK IO, the guest sending the packets, and then the guest receiving the packets.

Preparation for Defense
get Completion sign sheet for commitee, Course Form.
also need to book the room and send the abstract to get an announcement.
But the most important thing is to have the draft done.

gcc in debian

debian sarge with xfce in Xen express doesn't have gcc yet. To compile the bomb,
apt-get install gcc to get the gcc
apt-get install make to get the make

xen express problem

1 xen express cd
go to /client-install
rpm -ivh *.rpm
2 xenserver-client
input master-key
host IP and password

Xen Express only supports RedHat 4.1 and 4.4, RedHat5 needs VT supports, debian sarge, Windows XP sp2, Windows server 2003. No fedora or CentOS.

It can install from network(repository) or install from physical machine(p2v), but all limited to the few OSes.

Debian sarge is installed, apache install instruction:
1. apt-get update
to update the package sources
2. apt-get upgrade update
upgrade debian
3. apt-get install apache2
apache -v display the version
install webserver apache2.0.54,
The web directory is /var/www
apachectl restart restarting the webserver, or /etc/init.d/apache2 restart
3 apt-get install libapache2-mod-php4 php4-cli php4-common php4-cgi
The php version is 4.3.10

Set a static IP in debian
edit /etc/network/interfaces
auto eth0 lo
#iface eth0 inet dhcp
iface eth0 inet static
address xxx.xxx.xxx.xxx
netmask 255.255.255.0
network xxx.xxx.xxx.0
broadcast xxx.xxx.xxx.255
gateway xxx.xxx.xxx.1

update the interface
/etc/init.d/networking restart

cannot afford

It sounds like that ESX's network performance is not good at all.
The more weird thing is that when running network sending from one guest, the specweb clients are all broken. And the first time that I can catch a chance to evaluate the transmiting rate which is around 68Mbps.

first time, all four down because kernel network panic.
Second time, two broken, complaining exceeded max allowed overthink time of 20 sec. Please make sure that neither the server or client are overloaded. If server is overloaded, consider the nunber of SIMULTANEOUS_SESSIONS requested. If clients appear overloaded, add more clients. The rest two: one 6.71%good, 16.78%tolerant, Total errors: 66; the other 0%good, 0%tolerant, Total error: 207.

I ping google, it has sent 14 packets, 6 received, the packet lost 57% , the response time is 19.0ms.

In ESX, the support sessions are limited to 20, when with 5 sessions, it still have some degradation here. But when fewer guests running there, the degradation will be relieved in some extent.

The side effect is that when the network goes wrong, the client will also be broken. So it is hard to do anything to solve the problem.

The kernel also reports a network panic there. Lost Internet connections

All these may due to:
1. ESX performance overhead is too high
2. Bladecenter network card or infrastructure isn't well designed
3. Network driver is not well written
4. The other blade network usage has affected the testbed.

-------------------------------------------------------------------
Summary
guest sending packets 4 broken, none suvives
guest receving packets the misbehaved broken at the 3rd iteration, the degradation is 60%, the good ones survive with 40% degradations.
The receiving rate is 10Mbps
ping google while guest receiving packets, 7packets , 5 received, 28% packet loss, the response time is 20 ms /40ms