Christmas Load Balancers

Dear Santa,

All I want for Christmas is a load balancer that does not suck.  It should do the following:

  1. Terminate and offload SSL sessions at a rate of no less than 20,000/sec
  2. Provide in-memory content caching
  3. Support web logging, Apache-style (and despite what some hardware vendors thing, web logging is really important)
  4. Have some method of scaling capacity (particularly at the SSL layer)

I know this is a lot to ask for but I’ve been good this past year.

Comments closed | Trackbacks closed

Comments (6)

  1. Yusuf wrote:

    Have you looked at Zeus

    http://www.zeus.com/

    It’s software based so you can take advantage of Moore’s Law with faster hardware. Would be a good idea to see how this could be paired with the Sun T5440 which have onboard SSL ofloaders

    Tuesday, December 2, 2008 at 3:37 am #
  2. rwg wrote:

    Don’t forget “working IPv6 support that doesn’t cost thousands of dollars extra.”

    Tuesday, December 2, 2008 at 5:41 am #
  3. Todd Walker wrote:

    How about http://www.a10networks.com ? The A10 AX3100 is rated at 24,000 SSL connections per second. IPv6 included for free. RAM is utilized for caching – I think up to 50% of physical RAM.

    Tuesday, December 2, 2008 at 6:49 am #
  4. mrz wrote:

    @Yusuf: I am actually. Yesterday I moved https://versioncheck.addons.mozilla.org/ over to a three node ZXTM cluster.

    In general I like Zeus but I don’t like the no-CLI interface. I also wish their clustering method allowed me to advertise just one DNS record instead of multiple. That eats IP addresses fast.

    Consider a ten node ZXTM cluster and 10 websites, each of which requires its own IP address. That’s 100 IPs off the bat.
    I’m considering putting the ZXTM cluster behind some L4 load balancer like Cisco’s ACE or a Foundry ServerIron.

    Tuesday, December 2, 2008 at 9:44 am #
  5. mrz wrote:

    @Todd: We looked at A10 awhile back. Their caching support was very 1.0 (by default they didn’t cache 302s which we rely on). Worse was that since caching was just recently added, they didn’t have weblogging. Since some amount of requests would have been served directly out of the A10’s cache, the backend webservers would never have seen the log hit.

    I really really liked A10’s virtual-n-mode cluster though.

    I’m leaning more towards a multi-layer approach now with some L4 load balancer in front of something like Zeus’ ZXTM.

    Tuesday, December 2, 2008 at 9:48 am #
  6. Laura Thomson wrote:

    I’m all for decoupling. I’d vote for hardware LB in front of a farm of commodity cache boxes that could grow as needed (and be tuned/replaced etc). But what do I know ;)

    Tuesday, December 2, 2008 at 5:13 pm #