Skip to main content
M
Michael_Kennedy__12345_
Inspiring
April 20, 2010
Question

Performance Tuning

  • April 20, 2010
  • 2 replies
  • 2357 views

Hi,

I am looking for some recommendations on how to maximise the performance of FMS 3.5 for both live and vod applications.

All servers are 8 core, 12Gb RAM. CentSO 5.4 x64. 2 x 1Gb bonded NICs. All video is 500k. All FMS installs are pretty basic. No adapters, vhosts etc othern than default. SWF verification and domain locking enabled.

2 origins, with LVS director mode load balancing. Running the default live and vod applications. 6 edges configured with the address of the "Virtual" origin server, adn these servers are load balanced by another pair of LVS DR mode servers.

At the moment the servers seem to have difficulty delivering 150Mb or more. Adobe claim that they should be able to drive the NICs at full speed with little processor effort. How? I have searched for the server benchmarks but they do not appear to actually exist. Anybody ever see them? The only documentation I have seen if for very old Macromedia versions.

I probally need both server level and FMS configurations. If anybody can point me to some good documentation I would be grateful.

Thanks,

Michael

    This topic has been closed for replies.

    2 replies

    A
    Anonymous
    April 23, 2010

    Hi.

    Can you tell me about how much files do you serve from VOD types? Is there a kind of huge number of files?

    In the worst cenario, i'd have every piece of all VOD files been streamed. If i've got 1TB of VOD files, would be necessary to get distributed FLVCaches to suport 1TB of data to stream my entire repository? 1TB of MEMORY?!

    Is there to much files in your application? Are you having issues with kind of FLVCache segment Full?

    Can you and everybody else share knowledges on other post?

    http://forums.adobe.com/message/2755840#2755840

    Tks a lot!

    Everton

    T
    techeye
    Participating Frequently
    April 20, 2010

    >> Adobe claim that they should be able to drive the NICs at full speed with little processor effort. How?

    It's not hard, you just need to understand what network configuration is going to allow the server to work.

    Make sure that the bonded nic configuration is correct.  I don't know how well that will work

    with LVS, but if i recall correctly swtich assisted load balancing helped (mode 6?)

    and I was able to get over 2 Gb/s with 4 nics (note this was a load test with a

    single stream).  Also, make sure you are in full duplex mode and that the switches are not mangling the traffic.

    Also, LVS may be the bottleneck.  If you can hit just 1 server and get 950 mbps, more than likely LVS cant keep up.  Are you using NAT, TUN, or DR?  I am guessing if you take LVS out of the mix you will not have a problem.

    M
    Michael_Kennedy__12345_Author
    Inspiring
    April 21, 2010

    Hi,

    I do not think that the problem is network related, but I give info on the relevant parameters I use below in case they are relevant somehow. An identical network configuration on other applications works just fine.

    The load on the FMS servers is extremely high at times, with the fmsedge process running almost at 100% of processor utilisation for extended periods of time. In this state the server is effectively unusable. Did you achieve your 2Gb with an default configuration?

    My servers are substantially more powerful than those in the FMS whitepaper (http://www.adobe.com/products/flashmediaserver/pdfs/FlashMediaServer3_WhitePaper_ue.pdf). I have never been able to saturate a NIC. Are there OS level changes I should make to improve performance?

    Thanks,

    Michael

    LVS in in Direct Routing mode, so performance on LVS is not an issue. High load and poor performance occurs even in a non-LVS test environment.

    Bonding is using the IEEE 802.3ad Dynamic link aggregation protocol. Config is below. This is pretty much the same as using HP NIC Teaming for those in Windows world.

    cat /etc/sysconfig/network-scripts/ifcfg-bond0

    DEVICE=bond0

    BONDING_OPTS="mode=4 miimon=500"

    IPADDR=RealIP

    NETMASK=Subnet Mask

    NETWORK=Network Address

    BROADCAST=Broadcast Address

    ONBOOT=yes

    BOOTPROTO=none

    USERCTL=no

    cat /etc/sysconfig/network-scripts/ifcfg-eth0 (same for eth1)

    DEVICE=eth0

    USERCTL=no

    ONBOOT=yes

    MASTER=bond0

    SLAVE=yes

    BOOTPROTO=none

    IP Tables rules in place as per the LVS documentation

    T
    techeye
    Participating Frequently
    April 22, 2010

    >> I do not think that the problem is network related,

    Oh, but it really is   And how do i know this?  well you told me:

    >>  fmsedge process running almost at 100% of processor utilisation for extended periods of time.

    The fmsedge process handles all the network related stuff.  usually it is the core process that folks worry about the performance.

    Trust me.  Take LVS out of the mix (at least for testing-- my guess is that DR is causing all the requests to be tunneled through the RTMPT protocol, and thus the edge is taking a beating.), and if it still isn't working, your bonded mode is the issue (again, switch assisted balance was superior in my tests).  I've been through all these hoops before.

    Terms & ConditionsAccessibility statement
    Using the community Terms of use Privacy Cookie preferences Do not sell or share my personal information ad choicesAdChoices