[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <f2b55d220702181404l4c00e91eo94171aac56e80ce7@mail.gmail.com>
Date: Sun, 18 Feb 2007 14:04:24 -0800
From: "Michael K. Edwards" <medwards.linux@...il.com>
To: "Eric Dumazet" <dada1@...mosbay.com>
Cc: "Evgeniy Polyakov" <johnpol@....mipt.ru>, akepner@....com,
linux@...izon.com, davem@...emloft.net, netdev@...r.kernel.org,
bcrl@...ux.intel.com
Subject: Re: Extensible hashing and RCU
On 2/18/07, Michael K. Edwards <medwards.linux@...il.com> wrote:
> ... Much less vulnerable to cache eviction DDoS
> than a hash, because the hot connections get rotated up into non-leaf
> layers and get traversed enough to keep them in the LRU set.
Let me enlarge on this a bit. I used to work for a company that built
a custom firewall/VPN ASIC with all sorts of special sauce in it,
mostly focused on dealing with DDoS. Some really smart guys, some
really good technology, I hope they grab the brass ring someday. On
the scale they were dealing with, there's only one thing to do about
DDoS: bend over and take it. Provision enough memory bandwidth to
cold-cache every packet, every session lookup, and every
packet-processing-progress structure. Massively parallelize, spinlock
on on-chip SRAM, tune for the cold-cache case. If you can't afford to
do that -- and if you haven't designed your own chip, with separate
cache windows for each of these use cases, you can't, because they're
all retrograde loads for an LRU cache -- then a hash is not the right
answer. The interaction between resizing and RCU is just the canary
in the coal mine.
Cheers,
- Michael
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists