lkml.org 
[lkml]   [1997]   [Apr]   [30]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRe: Swap over network
Date
From: hm@seneca.muc.de (Harald Milz)
Subject: Re: Swap over network
To: linux-kernel@vger.rutgers.edu
Organization: Linux.DE
X-PGP-Fingerprint: 44 5B 4E 65 50 F5 99 2B 20 55 C1 B0 56 FC C4 00
X-Nospam: Unsolicited ads are unwanted and will be proofread for $199/hr.

John Wyszynski (wyszynsk@clark.net) wrote:
> that talks to some server process as if it was a device. Doing it that way
> it could be used for things other than swapping; including raid and striped
> drives which span machines.

Just to add $.02 to this. The way a well-known (at least to me) commercial
solution achieves this is a pseudo device driver which behaves like a
normal block device to the application or filesystem working on it. It
traps all the write() calls and actually writes the data across an
arbitrary IP network to a remote machine (using a proprietary UDP-based
protocol) and to its local underlying device. Data is received by the
remote pseudo device driver(s), written to its local disk and acknowledged
to the original initiator. Everything happens two-way.

The pseudo device is configured to point to the local and some remote
block devices and basically synchronizes data between them. To provide
for data consistency, a transaction log is kept (actually, it is a bit
map carrying dirty flags). Reads are always done locally, for performance
reasons. This way, all the mirroring is perfectly transparent on the
block device level.

How about that?

BTW Striping across machines is probably not very useful. If one machine
fails, data is completely lost. Mirroring, in turn, _is_ very useful, and
several huge customers worldwide are just in the process of installing the
said commercial product (no advertising intended). Data Centers are between
3 km and more than 100 km apart.

Keep in mind that with the described algorithm, data throughput
(application performance) pretty much depends on the network latency
(I/O transfer rates, important for database style apps) and the maximum
throughput in MB/s.

--
Harald Milz ----------- PGP Public Key: finger hm[AT]muc.de (Key ID 7ADC4839)

All bridge hands are equally likely, but some are more equally likely
than others.
-- Alan Truscott

\
 
 \ /
  Last update: 2005-03-22 13:39    [W:0.690 / U:0.024 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site