Re: [cf-bosh] proposed stemcell network performance tuning
there are two problems:toggle quoted message Show quoted text
1) certain load balancer versions and configurations have unexpected
behavior around tcp timestamps when there is a mix of windows and
non-windows clients (really a mix of timestamps and not-timestamps). the
result is the linux servers in cloud foundry sending resets long before
ports are exhausted.
2) to date, these parameters have been configured in an ad hoc fashion as
problems are encountered leading to a lot of variation in configuration
across the various cloud foundry components. the ad hoc solutions have in
some cases even been counter-productive: tcp_tw_recycle was previously
enabled on some components, exacerbating #1.
the changes amit proposes are not exhaustive, but rather conservative and
address exactly these two problems. other tuning might be beneficial to the
platform. such additional tuning is not required for these scenarios.
On Wed, Sep 30, 2015 at 6:05 PM, Joshua McKenty <jmckenty(a)pivotal.io> wrote:
Amit - I worry about changes to the former in the context of HTTP 1.0 and