[lkml]   [1996]   [Oct]   [18]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: "raw" block devices?
>On Thu, 17 Oct 1996, Theodore Y. Ts'o wrote:
>> [...] One of the ways how the SGI gets truely
>> awesome web server benchmarks is by doing zero-copy disk reads from the
>> filesystem, and zero-copy network writes. All you need the CPU for is
>> to set up the DMA's, lock down the memory pages, and stay out of the
>> way. :-)
>... and you need a network card that does hardware checksumming :(((

However, if you use a typical Ethernet LAN, then your network
card already does hardware checksumming. Of course, this is Ethernet
frame-level checksumming, not, IP- or TCP-level checksumming, but
there may be circumstances in which you can trust the LAN and
locally-connected hosts sufficiently to defer or avoid the IP- or
TCP-level checksums. (As an analogy, you sorta squint sideways at
your LAN-connected workstations and fileservers, and declare them all
components of a single computer system with the LAN as its internal
memory bus.)

Another option is to give each individual disk drive its
own IPv6 address, and have it transmit directly to the network,
bypassing the "system" CPU.

The Netstation project at ISI has researched some of these
issues; please see:

Craig Milo Rogers

 \ /
  Last update: 2005-03-22 13:37    [W:0.044 / U:6.104 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site