Thanks Will, checking it out now. Mine is on http://developer.amazonwebservices.com/connect/thread.jspa?threadID=45411&tstart=0, where some viable solutions are being discussed.
on our internal network, we used IPVS (userspace tool = ipvsadm) with ldirector. This has a number of schedulers for use with connection distribution. Read up on how to configure it (I would recommend using the SED scheduler with IPIP mode of redirection (not gate or masq). This will quite effectively give you equally distributed round-robin connections. Just ran a quick JMeter test against a HA IP address backed with ipvs + ldirector and works just the way you're looking for. On Wed, Apr 21, 2010 at 2:42 PM, William Oberman <ober...@gmail.com> wrote: > Brett: just so you know, I started a thread over in AWS's forums on this > issue: > > https://developer.amazonwebservices.com/connect/message.jspa?messageID=175221#175221 > No one has responded yet. If anything comes up, I'll reply here (and add > to > the wiki). > > will > > On Wed, Apr 21, 2010 at 8:37 AM, William Oberman <ober...@gmail.com> > wrote: > > > Brett: yes, I saw amazon recently added two levels of stickiness (load > > balancer, and application). I have both disabled. I was referring to > TCP > > stickiness, though my only attempt to test it was using HTTP over TCP. > I've > > mentally ruled out the new stickiness feature as the problem because I > don't > > get a consistent backend using different load balancer IPs from the pool > > (and, I'd assume that true stickiness would be global). But I did get > > stickiness on a per-load balancer IP basis. Even though my evidence is > all > > observation based, I'm in 100% agreement with "According to user reports > in > > other forum posts, clients from a single IP address will tend to be > > connected to the same back-end instance.". This is due to the fact that > at > > the exact moment my DNS resolved to a new LB from the pool, all > connections > > switched to a new backend instance every single time. All I want (and > need) > > is "anti-stickiness" at all levels, including on a individual load > balancer > > basis.... If there is no real solution, I'm left with switching > providers, > > or installing my own load balancer (HAProxy, nginx are early hits on load > > balancer + ec2, and I've configured HAProxy before). On a different > note, > > you mentioned you're doing HTTPS (which is load balanced at the TCP > level). > > As a FYI, you do know amazon isn't doing SSL termination, so you _can't_ > do > > proper sticky load balancing, and you'll lose the client IP address. > > > > Sebb: I'm fairly new to using Jmeter, but I'd be happy to try and figure > > out where in the wiki to add this information. The page on processing > > results was very helpful for me. > > > > will > > > > > > > > On Wed, Apr 21, 2010 at 6:11 AM, sebb <seb...@gmail.com> wrote: > > > >> May I suggest that the findings are added to the JMeter Wiki? > >> > >> This will make it easier to find, update and refer to later. > >> > >> On 21/04/2010, Brett Cave <brettc...@gmail.com> wrote: > >> > Hi William, > >> > > >> > Thanks for the feedback. Have 1 question: > >> > > >> > > >> > On Tue, Apr 20, 2010 at 10:12 PM, William Oberman <ober...@gmail.com > > > >> wrote: > >> > > >> > > > >> > > 2.) For a given ELB IP, there seems to be a static mapping of > client > >> IP <-> > >> > > backend instance. This is a slightly complicated statement that > >> assumes a > >> > > some knowledge of how amazon in general, and ELBs in particular, > >> work. If > >> > > it's still up, this page: > >> > > > >> > > >> > > >> > Are you referring to HTTP stickiness here, or did you find that client > >> IP > >> > <-> backend instance is mapped for TCP connections too? (have been > >> > discussing this on the forums, and not getting an answer to this). On > >> the > >> > 7th of April, Amazon introduced sticky HTTP sessions on ELB (check > the > >> > sticky forum post for more info - > >> > http://developer.amazonwebservices.com/connect/ann.jspa?annID=646). > >> This > >> > should result in each thread in a jmeter test plan going through to > the > >> same > >> > node if you have a cookie manager. Then again, if there is indeed a > >> static > >> > mapping of client IP to instance, you would need to use multiple > >> instances > >> > of jmeter-server with a central controller to effectively test load > >> > balancing > >> > > >> > One of the responses to my post contained the link below, which > states > >> > "According to user reports in other forum posts, clients from a > single > >> IP > >> > address will tend to be connected to the same back-end instance." but > i > >> was > >> > wondering if you have been able to verify this? Our scenario is > greatly > >> > affected by this characteristic of ELB, as our entire web app is > >> > HTTPS-based. > >> > > >> > > >> > > >> > > > >> > > > >> > http://www.shlomoswidler.com/2009/07/elastic-in-elastic-load-balancing-elb.html > >> > > has pretty much everything you need to know. > >> > > >> > >> --------------------------------------------------------------------- > >> To unsubscribe, e-mail: jmeter-user-unsubscr...@jakarta.apache.org > >> For additional commands, e-mail: jmeter-user-h...@jakarta.apache.org > >> > >> > > >
