DHCP Management Across a Diversified Network?
ET Admin writes "I work for a small Wireless ISP, where we are deploying new network hardware to allow for growth and contain broadcast traffic. All routing/switching equipment is Cisco. We use Linux stand-alone boxes and VMs (running on Win 2003 boxes). We have decided on a hybrid VLAN layout where we have certain VLANs limited by location, and other VLANs that are global across the network. And I want DHCP served across it all. Does anyone have experience with IPAM software that handles multiple DHCP servers? Our network is small so spending a couple grand is overkill at this point. Any recomendations to help me decide between serving DHCP from the Nix boxes, or from the Cisco gear? Knowing that a single DHCP server will handle from 100-500 hosts."
setup DHCP Relaying on the switches to forward/relay all dhcp request across the vlans and subnets to one (or two) dhcp servers
http://lmgtfy.com/?q=cisco+dhcp+relay&l=1
You can easily run hundreds of thousands of hosts off a single DHCP server. It is not cpu intensive particularly if you have a decent lease duration.
Just because you disagree doesn't make it offtopic or flamebait.
Someone in house here created it, and we use it across multiple vlans from a Gentoo box. It uses the ISC DHCPD server.
http://phpdhcpadmin.sourceforge.net
DHCP not used in IPV6 protocol
Seriously, do not use the Cisco gear to handle the DHCP. There are several ways to handle this, either have a system with an interface on all the networks, or setup your Cisco gear to forward the HDCP requests to the one subnet that does have your system.
With using Unix/Linux you can setup failover servers so that if one does not respond, the other will take over the requests and that way you will not lose DHCP across your entire network due to hardware/software issues on a single system. Go read up on dhcpd, it is not too difficult to understand, and is really probably your best low cost solution.
We were all warned a long time ago that MS products sucked, remember the Magic 8 Ball said, "Outlook not so good"
You need to use DHCP snooping to block rogue DHCP servers and block packets with forged MAC addresses on untrusted interfaces
You need IP source guard to block forced IP addresses on untrusted interfaces
Otherwise, you are at risk of DOS and/or compromise from malicious users, and at risk of instability and insanity caused by users who plug a rogue DHCP server (even something as simple as the LAN side of a Linksys gateway) into your gear.
DNSMasq. Nuff said.
"When life gives you lemons, don't make lemonade. Make life take the lemons back!" -- Cave Johnson
To everyone who tagged this "domyjobforme", I hope every single one of you gets the same response the next time you ask for help doing you job. At least this guy had the sense to say, "Hey, there's a community of people that contains a multitude of experts in many fields, I bet someone might have some good suggestions." And guess what else? Maybe some readers will find the suggestions helpful too. Ask Slashdot is for questions that the general community might find interesting and helpful, not just one guy. It's not just about the submitter, and it's certainly not about your need to be snide to those who recognize their shortcomings and try to expand their base of knowledge.
jX [ Make everything as simple as possible, but no simpler. - Einstein ]
Don't use your cisco gear to manage dhcp. It's better utilized doing it's primary function of routing and switching. Set up a Linux box to do dhcp. Setup multiple subnets and use the "ip helper-address" command on the interfaces of your Cisco router's to forward the dhcp requests to your Linux dhcp server. It's simple to do. Once upon a time I setup a 5000k node network doing that very same thing.
Hey, wait, VMware server's still an option for production servers. Several years ago, it was a commercial product called VMware GSX server.
"Small wireless ISP" doesn't exactly strike me as the type of user, who would be deploying an Oracle RAC cluster with a load of 10k transactions per second, and an Exchange 2007 server with 5000 mailboxes, processing 10 messages per second.
GSX was the version for production servers in a small environment. ESX was the high-end uber-expensive version for running massive numbers of servers on a dedicated host in a large environment.
Server hardware in common use has gotten a lot better, much more powerful, since then. And VMware Server is no worse than GSX.
If your workload is suitable for that type of virtualization, GSX should be okay.
Yeah, ESX is a lot better, can handle many more VMs, and can virtualize many high-end workloads effectively that weren't even VM-suitable under GSX/VMware server.
ESXi is less mature, and probably not as suitable as ESX.
I don't know enough about your environment but hopefully you know that that isn't a possibility across Layer 3 devices (and when I say VLAN's, I assume that you are talking about an IP segment and not just a VLAN number). That said the "ip dhcp helper" or DHCP relay I think is what you are looking for. This way you can have 1 DHCP server serving numerous VLAN's or L3 IP segments. If you have more specific questions feel free to reach out to me.
Carl Fugate
carl@iprouteradmin.com
BLOG: www.iprouteradmin.com
Router Lab: www.onlinerouterlab.com
Using one or two of your Win2003 boxes, create multiple DHCP scopes for your multiple networks/subnets. Then just use the "ip helper-address" on your cisco gear to allow the DHCP requests to make it to your servers. Done. I do this at my company with 50+ VLANs.
Cost = $0.
"A plan fiendishly clever in its intricacies"- Homer Simpson
Cisco make (or at least did make some time ago) a DHCP server (Cisco Network Registrar) based on Windows that does handle option 82. So you do not have to run DHCP on IOS, you can relay back to a central server. I have even been able to "script" CNR by sending command line commands to administer scopes (yes, thank god it has a command line). But in all honesty, it's far easier to automate the configuration of a standard linux or *BSD dhcpd.
Nullius in verba
for 100-500 leases, i would set it up on the cisco boxes.... also ensures the zero cost approach which always makes management happy.
I have to ask, who will be monitoring and supporting this architecture?
ESXi might be less mature, but it's free! Works pretty flawlessly and it's so damn similar to ESX it's not funny. Major differences are you can't tie a VM to a serial port and no real *nix console (there is a console, but it's f'ing limited) all CLI stuff is replaced with the Remote CLI, which I hear is pretty powerful, but I've not had the need to use it yet.
# cat
Damn, my RAM is full of cats. MEOW!!
Carnegie Mellon's NetReg is an open source system that provides a pretty complete IP Address Management toolset, including management of DNS & DHCP configurations for ISC bind/dhcpd. It can manage ISC dhcpd's failover configuration, and multiple server groups, etc.
Rather then just repeating what I've said before when the subject of IP Address Management came up on slashdot, I'll just link to it.
Note: While the project has been pretty quiet for quite some time now, thats mostly because its the system is very stable and there hasn't been a lot of major new development in the last couple of years. I used to be one of the core developers of the system before I moved on to another job, but its still in active use by many sites.
There are more, depending on the exact setup you're deploying and the level of complexity. (DHCP Option 82 for example)
I would have a look at http://www.weird-solutions.com/
They produce some cutting edge DHCP and provisioning software for amongst others the ISP market. Furthermore their staff are incredibly knowledgeable.
Logging is the main reason I like our current dhcpd setup. I tried the dhcp debugs on the cisco gear and didnt get much.
Fixed your title for you.
My opinion of ESXi is it's great, but you really need VMotion with it, because (among other things) ESXi seems more prone to instability of the management interface, mainly because it has fewer allocated resources.
Well, i've seen the ESXi "thin management interface" running out of memory, such that the VI client could barely connect, and such, it's not fun, and requires a reboot of the machine, since there's no proper console to SSH in on. I've had unique instability issues with ESXi. And also been hit by PSOD (Pink Screens of Death) in ESXi, but not when testing ESX.
ESXi doesn't provide a support remote management solution other than the graphical VI client. I realize there's a hack that lets you get access to a busybox shell, but VMware may close this at any time, in a later critical fix, i.e. it's not something to rely upon.
You can't really buy support from VMware and use it, they'll blame your problems on you using an unsupported hack.
The remote management API and the Remote CLI's access is restricted to read-only operations, unless you buy at least a Foundation level license.
i.e. It's free to use the remote CLI in read-only mode, but as soon as you want to do something like clone a VMDK file, or power on a VM, the free version doesn't allow it.
The secret is VMware accidentally "enabled full write access" in ESXi 3.5 Update 3. (But they realized their mistake and made it read-only again, re-imposed the restrictions in the next version)
There are some other bugs in update 3, mainly security bugs, and issues that effect iSCSI and SAN users.
But if you utilize direct attached storage, and only run trusted code in your VMs, installing the old version of ESXi and using 3.5 may be an option, for having at least some minimal amount of proper remote management and scriptability.
Thanks, I hadn't run into a lot of those. Good to know! Great for a home dev environment though. Before I moved house, my box had been running Nearly a year straight. Uptime not so great at the moment, long power failure and I've done a few upgrades!
# cat
Damn, my RAM is full of cats. MEOW!!
I'll throw out my solution.
:)
As many people here have suggested, ISC DHCP server has no trouble with this and can handle many subnets and pool combinations from one or more servers. Then with the combination of ip helper-address on Cisco platforms you can control which server(s) handle the network. Throw DHCP-Failover into the mix and make it redundant.
To manage all this I'd suggest OpenNetAdmin. It is geared to manage as any IPAM would, your address space. It can also be instructed to manage multiple DHCP servers in whatever combination you need. Then those servers simply extract their specific configuration from the database. It should have no issue scaling to several hundred distributed DHCP servers if needed. It will all however be managed easily via the centralized WEB/CLI interface. Opennetadmin will also keep track of your vlan information as well.
I would personally avoid running DHCP on the cisco devices, but thats just me.
Hope that helps. Again, head to http://opennetadmin.com/ and see if that works for you!
Thanks