-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/24719/#review50746
-----------------------------------------------------------



docs/network-monitoring.md
<https://reviews.apache.org/r/24719/#comment88614>

    Perhaps make explicit that they can reach other containers, including on 
the same host, and that they can still reach services running on the localhost 
of the host <--- This is important because you wouldn't usually expect this if 
you had container's running with separate network stacks.



docs/network-monitoring.md
<https://reviews.apache.org/r/24719/#comment88612>

    Same public IP and same MAC?
    
    s/communicate with external world/make connections with other hosts/?
    
    Are you going to mention about multiple ICMP/ARP replies anywhere?



docs/network-monitoring.md
<https://reviews.apache.org/r/24719/#comment88601>

    s/By default, Linux uses ports [] for ephemeral ports./The default 
ephemeral port range on Linux is []/
    
    Is [32768, 61000] the default range across all distributions?
    
    s/used by the container as ephemeral ports/used as the ephemeral port range 
for the container's network stack/
    
    Can you add a sentence that these ports are *directly mapped* into the 
container's port range, hence the naming network/port_mapping.
    
    This ties into each container having its own 
/proc/sys/net/ipv4/ip_local_port_range.
    
    s/need to make sure/need to ensure/
    
    s/enforce the squeeze/ensure there are no connections using ports outside 
the new ephemeral range./



docs/network-monitoring.md
<https://reviews.apache.org/r/24719/#comment88602>

    Is it not more common to just use `echo "" > /path` while running as root? 
People may not be familiar with the tee trick.
    
    I think you should explain the reasoning and potential impact this change 
has. Clarify that this reduces the ephemeral range for any process that isn't 
is a container and that it limits the number of connections based on the 
5-tuple and includes TIME_WAIT so highlight that this should be evaluated for 
scenarios which have a high number/churn of connections to services.



docs/network-monitoring.md
<https://reviews.apache.org/r/24719/#comment88605>

    Explicitly state that the maximum number of containers on the slave will be 
limited by approximately |ephemeral_ports|/ephemeral_ports_per_container, 
subject to alignment etc.
    
    E.g., for these numbers the slave is limited to 24 containers. This is an 
important limitation that should be made very explicit.



docs/network-monitoring.md
<https://reviews.apache.org/r/24719/#comment88606>

    The number of ephemeral ports is not power 2 *aligned*, it's just a power 
of 2.
    
    Give an example here for power of two size to guide people, e.g., 512, 1024 
or 2048.
    
    s/what/which/
    
    Explain that non power-2 sized will have some performance impact for 
handling packets.



docs/network-monitoring.md
<https://reviews.apache.org/r/24719/#comment88608>

    Can you clean up the floating point madness here? e.g., cpus_limit: 0.35. 
It's not useful and is just distracting.


- Ian Downes


On Aug. 14, 2014, 5:13 p.m., Jie Yu wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/24719/
> -----------------------------------------------------------
> 
> (Updated Aug. 14, 2014, 5:13 p.m.)
> 
> 
> Review request for mesos, Chi Zhang, Ian Downes, Vinod Kone, and Cong Wang.
> 
> 
> Repository: mesos-git
> 
> 
> Description
> -------
> 
> See summary
> 
> 
> Diffs
> -----
> 
>   docs/network-monitoring.md PRE-CREATION 
> 
> Diff: https://reviews.apache.org/r/24719/diff/
> 
> 
> Testing
> -------
> 
> checked the markdown syntax
> 
> 
> Thanks,
> 
> Jie Yu
> 
>

Reply via email to