CEPH: TCP Performance Tuning

ethernet-cable-fade

Below are a few TCP tunables that I ran into when looking into TCP performance tuning for CEPH.

Note that there are two separate sections for 10GE connectivity, so you will want to test with both to find what works best for your environment.

To implement, we just add what is below to /etc/sysctl.d/99-sysctl.conf and run “sysctl -p“. Changes are persistent across reboots. Ideally these TCP tunables should be deployed to all CEPH nodes (OSD most importantly).

## Increase Linux autotuning TCP buffer limits
## Set max to 16MB (16777216) for 1GE
## 32MB (33554432) or 54MB (56623104) for 10GE

# 1GE/16MB (16777216)
#net.core.rmem_max = 16777216
#net.core.wmem_max = 16777216
#net.core.rmem_default = 16777216
#net.core.wmem_default = 16777216
#net.core.optmem_max = 40960
#net.ipv4.tcp_rmem = 4096 87380 16777216
#net.ipv4.tcp_wmem = 4096 65536 16777216

# 10GE/32MB (33554432)
#net.core.rmem_max = 33554432
#net.core.wmem_max = 33554432
#net.core.rmem_default = 33554432
#net.core.wmem_default = 33554432
#net.core.optmem_max = 40960
#net.ipv4.tcp_rmem = 4096 87380 33554432
#net.ipv4.tcp_wmem = 4096 65536 33554432

# 10GB/54MB (56623104)
net.core.rmem_max = 56623104
net.core.wmem_max = 56623104
net.core.rmem_default = 56623104
net.core.wmem_default = 56623104
net.core.optmem_max = 40960
net.ipv4.tcp_rmem = 4096 87380 56623104
net.ipv4.tcp_wmem = 4096 65536 56623104


## Increase number of incoming connections. The value can be raised to bursts of request, default is 128
net.core.somaxconn = 1024

## Increase number of incoming connections backlog, default is 1000
net.core.netdev_max_backlog = 50000

##  Maximum number of remembered connection requests, default is 128
net.ipv4.tcp_max_syn_backlog = 30000

## Increase the tcp-time-wait buckets pool size to prevent simple DOS attacks, default is 8192
net.ipv4.tcp_max_tw_buckets = 2000000

# Recycle and Reuse TIME_WAIT sockets faster, default is 0 for both
net.ipv4.tcp_tw_recycle = 1
net.ipv4.tcp_tw_reuse = 1

## Decrease TIME_WAIT seconds, default is 30 seconds
net.ipv4.tcp_fin_timeout = 10
 
## Tells the system whether it should start at the default window size only for TCP connections
## that have been idle for too long, default is 1
net.ipv4.tcp_slow_start_after_idle = 0
 
#If your servers talk UDP, also up these limits, default is 4096
net.ipv4.udp_rmem_min = 8192
net.ipv4.udp_wmem_min = 8192
 
## Disable source redirects
## Default is 1
net.ipv4.conf.all.send_redirects = 0
net.ipv4.conf.all.accept_redirects = 0

## Disable source routing, default is 0
net.ipv4.conf.all.accept_source_route = 0

Reference here

How to Create Non-Routable Isolated (but not Private) Vlans on a Cisco Catalyst Layer 3 Switch

data_sheet_c78-530976-1

First off let’s start out by saying that Isolated VLANs and Private VLANs are two completely different things… they are not at all the same. To a network administrator, this should make perfect sense. However, a Server or Virtualization Administrator may or may not know the different. Because of this, I hear many non-network Administrators toss around the term “Private VLAN“, when they actually mean to say “Isolated Vlan“, or more specifically what they are referring to is a “Non-Routable” VLAN.

What’s confusing is that the networks that we plan to use over our newly commissioned Non-Routable VLANs can correctly be refereed to as Private networks. They are private because no traffic can get in our out without a direct lP link to this network. However the VLANs themselves are not private, just isolated, or non-routable.

I believe that you can see where the confusion comes from.

So allow me to provide a bit of context before we go any further.

In my specific case, I need to create what are commonly (however incorrectly) referred to as Private VLANs to act as a back-end network for an OpenStack deployment. I cannot tell you how often I have heard someone make this mistake. This new VLAN, or network, needs to remain isolated from the outside world, meaning that it does not need to be able to route to any other network, or out to the internet. Rather, this new VLAN needs to send isolated traffic back and forth between network nodes deployed as part of my OpenStack Deployment. What I am describing here is not a “Private VLAN”, it is a “Non-Routable”, or “Isolated VLAN”

So please let’s make sure that we are using the correct terms.

So here is how you do it.

In my case I want to create two isolated VLANS for isolated traffic between my OpenStack nodes. Note that I am using nested virtualization, so my OpenStack nodes are themselves VMs.

First lets create what I will refer to as NR-1 (non-routable-1). We will use the VLAN id 666 as its easy to remember.

s3560#conf t
Enter configuration commands, one per line. End with CNTL/Z.
s3560(config)#vlan 666
s3560(config-vlan)#name NR-1
s3560(config-vlan)#end

Now lets create what I will refer to as NR-2. (non-routable-2)

s3560#conf t
Enter configuration commands, one per line. End with CNTL/Z.
s3560(config)#vlan 667
s3560(config-vlan)#name NR-2
s3560(config-vlan)#end

How lets check out our vlans, starting with 666

s3560#show vlan id 666

VLAN Name Status Ports
—- ——————————– ——— ——————————-
666 NR-1 active

…trunc…

Now let’s take a look at 667

s3560#show vlan id 667

VLAN Name Status Ports
—- ——————————– ——— ——————————-
667 NR-2 active

…trunc…

Note, that if I wanted to make these VLANs routable, I would need to add a layer3 interface. We are obviously not going to do that here.

Now lets add these new VLANs to our existing virtualization server trunks. We are going to do this to a range of interfaces to save time. Note that I was already allowing VLANS 101-104 and 192 on these trunks.

s3560#conf t
Enter configuration commands, one per line. End with CNTL/Z.
s3560(config)#interface range GigabitEthernet0/15 -18
s3560(config-if-range)#switchport trunk allowed vlan 101-104,192,666,667
s3560(config-if-range)#end

Now don’t forget to save our config.

s3560#copy run start
Destination filename [startup-config]?
Building configuration…
[OK]
0 bytes copied in 1.443 secs (0 bytes/sec)

Configure Syslog Logging Levels on the Asus RT-AC66U Router


4614_WizardStressToy_1

So here is a quick little one that I figured out the other day. Having just setup a Splunk server at home I wanted to make sure that I was not going to hit the data limit of 500mb a day for the free version of Splunk. I figured out pretty fast that my ASUS RT-AC66U was a very chatty-cathy when it came to syslog… sending me all sorts of very raw data that I was, at least at first, not so sure I was interested in indexing. So I hit the cli and started poking around.

First off, before we jump in, let’s make sure that we are all on the same page. First thing to note is that I am running the custom Merlin firmware, however that doubt that the stock firmware is much different. Second, let’s make sure that we all know how to configure syslog on our Asus.

To setup forwarding syslog to a remote syslog server, you first client on “Administration” in the “Advanced Settings” panel on the left. Then select the “System” tab near the top of the page. Scroll down to “Miscellaneous”. This section is shown below. Enter the IP address of your syslog server (or Splunk server in this case) in the “Remote Log Server” field.

syslog_asus

Now lets get down to the business of adjusting our logging level. First you need to ssh into your router.

Note that it appears that by default the log level is set to 7.

admin@RT-AC66U: # nvram show | grep log_level
log_level=7

Now before you get too excited, I am actually not sure that the main log level adheres to rfc5424. I have yet to find any published documentation from Asus to confirm this. However, according to this guy’s blog, this configuration might be a bit less chatty. Note that there are a few additional settings here which you can play around with. With these settings, I am assuming that 1 is on, and 0 if off. I am still experimenting.

admin@RT-AC66U: # nvram set log_level=2
admin@RT-AC66U: # nvram set log_enable=1
admin@RT-AC66U: # nvram set log_rejected=1
admin@RT-AC66U: # nvram set log_dropped=1
admin@RT-AC66U: # nvram set log_accepted=0

Now lets save our change and reboot

admin@RT-AC66U: # nvram commit
admin@RT-AC66U: # reboot

Note that there also is a vpn_loglevel=3 setting that can be configured via nvram. This setting might be useful to those running a VPN server on their router.

Redhat 6 Minimal Kickstart Configuration with VMware Tools and Puppet Agent Install

smartaHere is my small, crude, little Kickstart configuration and post install script that I have up and running in my lab at home. Don’t expect to find anything too fancy here, as this Kickstart was purposefully built to be small and to the point. Here, the point was to spin up a VM, run through a basic install of CentOS/Redhat Linux,  and install VMware Tools along with a Puppet agent.

Note that this post assumes that you have a working Kickstart server.

First lets take a look at our kickstart file, CentOS-6.6-x86_64-minimal.ks

The section directly below kicks off our kickstart ks file. Here we set our root password (no that’s not my hash) and setup our network interface for DHCP. We do a tiny bit of disk partitioning, and setup very simple LVM. Then we choose our packages. As you can see my package list is not at all fancy, I just want to make sure that I have pretty much every package that might need for a lab VM.

# Kickstart file for RHEL 6 Minimal
# Small Disk

text
install
url --url=http://10.1.0.106/ks/loop/CentOS-6.6-x86_64-bin-DVD1
lang en_US.UTF-8
keyboard us
network --onboot yes --device eth0 --bootproto dhcp --noipv6
rootpw --iscrypted $6$X/4YYZPN$4Sv.khxXms8N8vRssR/Vl35w/m80FF5P6p7aX0D7EFfD9p734F6tU4kXdcSCoOjPiXLrVxqfKxxxxxxxxxxxq5551
firewall --disabled
authconfig --enableshadow --passalgo=sha512
selinux --permissive
timezone America/New_York

# Disk
bootloader --location=mbr --driveorder=sda --append="crashkernel=auto rhgb"
zerombr
clearpart --all --drives=sda
part pv.1 --grow --size=1
part /boot --fstype=ext4 --size=1024
volgroup VolGroup pv.1
logvol / --fstype=ext4 --name=lv_root --vgname=VolGroup --size=1024 --grow
logvol swap --name=lv_swap --vgname=VolGroup --size=1024

#Network
network --device=eth0 --bootproto=dhcp --nameserver=10.1.0.110

# Package Selection
%packages --nobase --excludedocs
@Base
@core
kernel-headers
wget
perl
sysstat
bind-utils
tcpdump

Now let me pause to point out the section below. This is the %pre script that I am using to prompt me for the VM hostname before the install begins. The hostname needs to be set before you install puppet on the VM, otherwise you are going to have to recreate your puppet certificates after you set properly set your hostname post install and reboot.

%pre --log=/root/ks_pre.log
#change to tty6 to get input
chvt 6
exec </dev/tty6 > /dev/tty6

#Prompt for hostname
echo "What is my hostname?"
read NAME
echo "NETWORKING=yes
HOSTNAME=${NAME}" > network
chvt 1

Now we run a simple post install, along with a custom post install script. It is this script that will install Vmware tools and Puppet. Myself, I prefer keeping most of my code out of the actual Kickstart ks file, however you can always jam all your code into it if you like. You will just need to validate your syntax first, as I have not tested my config this way.

%post --nochroot
# bring in hostname collected from %pre, then source it
cp network /mnt/sysimage/etc/sysconfig/network
. /mnt/sysimage/etc/sysconfig/network
# force hostname change
/mnt/sysimage/bin/hostname $HOSTNAME
#Post Install
%post --log=/root/ks-post.log
cd /root
echo "Getting the post install script - if this takes a long time check network or path"
wget http://10.1.0.106/ks/scripts/centos-6-postinstall.bash
echo "Running the post install script"
/bin/bash centos-6-postinstall.bash

Ok, so below is the post install script that I am calling in the section above. After a quick modification of my hosts file, I pull down the Puppet installer from my local Puppet server. Next we install the open source VMware tools packages, after creating the required yum repofile.

 

#!/bin/bash
#centos-6-postinstall.bash

#Switch to the 6th console and redirect all i/o
exec < /dev/tty6 > /dev/tty6 2> /dev/tty6
chvt 6

# Lets make sure we know who the puppet server is before we get too far
echo "Adding hosts entry for puppet master"
echo "10.1.0.115 puppet puppet.lab.localdomain" >> /etc/hosts

## Update Via Yum - not doing this for now in order to save time
#yum -y update
# Install puppet from local puppet master
echo "Downloading and running Puppet installer"
curl -k https://10.1.0.115:8140/packages/current/install.bash | sudo bash
#Install Open Source VMware Tools
rpm --import http://packages.vmware.com/tools/keys/VMWARE-PACKAGING-GPG-DSA-KEY.pub
rpm --import http://packages.vmware.com/tools/keys/VMWARE-PACKAGING-GPG-RSA-KEY.pub

echo -e "[vmware-tools]\nname=VMware Tools\nbaseurl=http://packages.vmware.com/tools/esx/5.1latest/rhel6/$HOSTTYPE\nenabled=1\ngpgcheck=1" > /etc/yum.repos.d/vmware-tools.repo

echo "Installing Vmware Tools"
yum -y install vmware-tools-esx-nox

#Minor grub.conf modifications
sed -i 's/rhgb quiet//' /boot/grub/grub.conf
sed -i 's/hiddenmenu//' /boot/grub/grub.conf
sed -i 's/timeout=5/timeout=10/' /boot/grub/grub.conf

#Kick off first puppet run, for some reason I think you might need to do this twice.
sleep 5
echo "Running Puppet for the first time"
puppet agent --test
puppet agent --test

#Tell us we have reached the end
echo "We have reached the end of the post-install script"

A couple of additional details to note about the post install script above. I like to modify the grub.conf so that I unhide the menu and increase the time out. I also like to make sure that we disable the Redhat graphical boot screen… I want to make sure its easy to catch any errors or miss-configurations in my kickstarts.
 

Vmware CapacityIQ Unregister a Vcenter Server via the CLI

Vmware-logoOk two in a row, now thats strange. Anyway.. I was working on a rebuild and re-ip of my Vcenter Server as I was in the process of changing over to the Vcenter appliance, when i realized that I was going to have to lay hands on a lot of tools and change them to point/connect to my new Vcenter Server instance. Anyway. I started of with VCOPS, and I was surprisingly sucessful. So I figured I would move on to Capacity IQ.

Again, being the Linux guy that I am, I hit up the cli and stated fiddling around. I stumbled across the ciq-admin command and was able to find the following

[ciqadmin@ciq ~]$ ciq-admin register status
vCenter Server Registration Status = Registered
vCenter Server address             = https://vc00.lab.localdomain/sdk
vCenter Server user                = LAB\user

 

Hey ok, that looks good so far. Lets see if we can unregister.

[ciqadmin@ciq ~]$ ciq-admin unregister
Connection to VC failed. Disconnected CIQ, but could not remove CIQ extension from VC.

 

Well crap, you will never be able to connect to my original Vcenter Server… it's dead and its not coming back. Luckily it appears that we can force the registration of a new Vcenter Server

[ciqadmin@ciq ~]$ ciq-admin  register –vc-server vc01.lab.localdomain –force  –user root –password mypasswd
Stopping CIQ :                                             [  OK  ]
Starting CIQ :                                             [  OK  ]
ciq-watchdog is now enabled

 

Now head back on over and get a gander at the webUI. Still shows the old Vcenter Server? Just logout and log back in again and life should be good.

 

Related articles

Vmware VCenter Virtual Appliance – Death to Windows.. I think.
ESX 5: How to Power On A Virtual Machine from the Command Line
VMware vCenter Operations Manager Essentials – Introduction to vCenter Operations Manager
vCenter Operations Manager: VMware's move into cloud monitoring?

Vmware Vcenter Operations Manager Unregister a Vcenter Server via the CLI

Vmware-workstation-17-535x535So my Windows based Vmware Vcenter Server went belly up again. Something to do with the SSO database not starting. Not being a lover of Windows I decided to give the Vcenter Server Appliance a shot. Install was great and I am kicking myself a bit as to why I spent so much time fighting with Windows. My new Vcenter Server, which has a different IP address then the original Windows box (might make a great Veeam server) was not registered with Vcenter Operations Manager. I was not prepared to reinstall that thing again. So I needed to figure out how to manually unregister a Vcenter instance and add register another one in its place.

Since VCOPs runs Linux, I decided to ssh into the server and see if I could figure it out. First thing I found that I needed to do was figure out the registered Vcenter Server name and Vcenter Name (whatever that's supposed to be) I was able to do this using the vcops-admin command.

admin@vcops:~> vcops-admin summary

 

This command output a bunch of stuff, but the important bits for this task are below.

Registration Details
——————–
vCenter Server address  = https://vc00.lab.localdomain/sdk
vCenter Server name = Lab Vcenter

 

So now we need to unregister the sucker above. Note that this command takes a bit to unregister

admin@vcops:~> vcops-admin unregister –vc-name Lab\ Vcenter –vc-server https://vc00.lab.localdomain/sdk –user LAB\\userid –password mypassword–force

vCenter Server unregister = success

 

So flip on over to your browser and log in. The unregister process will cause the webUI to reload, so if you were already logged in you will find that you still need to log in again.

Now you can register your new Vcenter Server via the webUI.

Related articles

Vmware VCenter Virtual Appliance – Death to Windows.. I think.
ESX 5: How to Power On A Virtual Machine from the Command Line
vCenter Operations Manager: VMware's move into cloud monitoring?
VCOPs – VMware's Move into Cloud Monitoring

How to Add and Delete Persistant Routes in AIX

Chatter-phone-300x234Wow, AIX does not like to make anything easy. Nor do they like to make things intuitive. Need to remove a route from AIX, well get ready to have one command to temporariliy remove a route and another command to remove the route for good. Same goes with adding a route. "Quit your bitching, and use Smitty", you say? Well smitty does not make anything any easier, especially since the UI likes to show you fields that you does not necessarily need you to use. Oh, plus they let you type in them. Asking me for a netmask when adding a static route does not seem like a crazy request to me? But jokes on you, you were not supposed to type anything there.

Anyway, I was tasked with cleaning up a few bad routes that were added to a handful of servers. Note that I was not interested in adding the routes temporarily before I added them as persistent routes. These were simple one liners that I felt very comfortable adding to the ODM right out the gate.

So first we need to check the ODM for the routes that we need to remove. In this example we want to delete the route fro 10.11.1.56, so lets find just that route. Note all of these are host routes.

# lsattr -El inet0 | grep 10.11.1.56
route host,-hopcount,0,,-if,en1,,,,-static,10.11.1.56,10.22.13.1 Route True

 

Now lets delete the route above

# chdev -l inet0 -a delroute=10.11.1.56,10.22.13.1

 

We have a second route to delete so lets kill that one too

# chdev -l inet0 -a delroute=10.11.1.57,10.22.13.1

 

Now lets add the correct routes – the format is as shown below

chdev -l inet0 -a route=host,-hopcount,1,-netmask,netmask,network,gateway

 

Now on a few of the servers I was working on I had to remove network routes as well as host routes. You need to know that the syntax is anoyingly different adding network routes. In the example below I first need to remove the network route.

chdev -l inet0 -a delroute="net,-hopcount,0,,,,,,-static,10.11.1.56,10.11.1.254"

 

Now lets add our new and correct route

chdev -l inet0 -a route="host,-hopcount,0,,,,,,-static,10.11.1.56,172.30.150.190"

Related articles

Basic AIX Performance Troubleshooting Commands
How to Add Multiple Routes in Linux Using ip Command Examples
PowerPath, AIX 5.3, and LTG: Solutions