Forum Discussion

Logan_Ramirez_5's avatar
Logan_Ramirez_5
Icon for Nimbostratus rankNimbostratus
May 26, 2009

current connection question on get_global_stats

so I'm pulling the SystemStatistics.get_global_statistics() variable and posting the STATISTIC_CLIENT_SIDE_CURRENT_CONNECTIONS value to a screen and it's around 300,000 (just shy).

 

 

(this value hold true when comparing it to the Active Connection graph in the Overview --> All --> Performance GUI)

 

 

Ok, so the obvious question, then is this:

 

 

What is the maximum client side (and/or server side) connections an LTM 1500 is capable of supporting?

 

 

Or as my programming logic goes, at what connection threshold should I change my screen to red!?

 

 

The tech specs online talk about throughput (500Mbps), but I can't find any docs on 'connections' and then, I suppose, not really sure if that value is relevant...but surely there must be a relationship between connections and memory, throughput, etc!?

 

 

any thoughts?
dev
devops
iControl

5 Replies

Sort By
  • Joe_Pruitt's avatar
    Joe_Pruitt
    May 26, 2009
    Connections/Sec is a tough one because it's totally dependent on what the connections are doing. Basically you need to determine how much data a single connection is taking up and divide the 500mb/s by that number and you can come up with a Connection/sec value. I did find a reference in the "BIG-IP 1600/1500 Platforms: Performance Per Watt" datasheet

     

     

    http://www.f5.com/pdf/products/hardware-ppw-comparison-1600.pdf

     

    Click here

     

     

     

    that gives a number of 22k but I'm not sure if that number just a baseline as to what you get with the 1600 relative to the 1500 with those values.

     

     

    Wish I could give you something more concrete. I'll ask around and see if I can find you some relative numbers based on average application connection requests.

     

     

    -Joe
  • L4L7_53191's avatar
    L4L7_53191
    Icon for Nimbostratus rankNimbostratus
    May 26, 2009
    At pure layer 4 a 1500 will handle somewhere around 4 million concurrent connections, but in practice most people don't run in pure layer 4 mode so this number is subject to change because not all traffic is created equal...some of the folks in the application deliver space ignore this fact (forunately F5 isn't one of them - look at the performance report here: http://www.f5.com/reports/f5-performance-report.pdf).

     

     

    How utilized is this system currently (memory/cpu, etc)?

     

     

    -Matt
  • Mike_Lowell_108's avatar
    Mike_Lowell_108
    Historic F5 Account
    May 26, 2009
    I encourage monitoring CPU, memory, and throughput as your primary metrics. These are things you can run out of. A high connections per second value is a likely explanation for why your CPU usage is high, but it's only indirect -- there isn't a specific conn/s limit: you don't run out of conn/s, you run out of CPU. Similarly, a high conurrent connections value would correlate with a high memory usage.

     

     

    Reasonable values for conn/s and conc. conns are tough to suggest generically, so I'd encourage you to primarily monitor CPU/mem/throughput, and secondarily conn/s, concurrent conns to give you an idea of what level of conn/s and concurrency correpsond to a given level of CPU/memory/throughput.
  • Logan_Ramirez_5's avatar
    Logan_Ramirez_5
    Icon for Nimbostratus rankNimbostratus
    May 27, 2009
    Yes! Wait, I mean 'Yes' seconding the 'thanks for the help' not 'Yes' the answers were better (though, ok, ok, yes, yes, they were, in fact, at the very least marginally better! ha!).

     

     

    this is a very useful information. thank you.

     

     

    to answer the questions from yesterday, the system CPU is below 10 (rarely spikes up to 30 let alone 50) and the Mem shows 138 of 430 (which is interesting since the base package as 768 and I'm pretty sure we have a 1G).

     

     

    I will work on building monitors for CPU, Mem, and Throughput!

     

    Muchas gracias.