From: Jim Starkey
Date: November 4 2008 11:31pm
Subject: Re: Network vs. Disk
List-Archive: http://lists.mysql.com/falcon/138
Message-Id: <4910DB37.8090102@nimbusdb.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Brian Aker wrote:
> Hi!
>
> I will riddle you back!
>
> On Nov 4, 2008, at 10:50 PM, Jim Starkey wrote:
>
>> Cheap servers (quad core Intel) can send 39,.000 messages through a 
>> switch per second without breaking a sweat (measured between WinXP 
>> and 64 bit Linux both running Q9450 quad-core Intel processors @ 2.66 
>> GHz).  A disk with an average access time of 6 milliseconds can 
>> sustain 166 operations per second.
>>
>> What does this suggest about the future of database systems built 
>> around disks?
>
> If memory that is attached attached the processor is no longer 
> volatile how ludicrous is it to develop using a paradigm of "disk" and 
> "main memory"?
>
Not ludicrous at all.  Let's see:

   1. Redundancy for availability
   2. Redundancy for scalability
   3. Redundancy for disaster recovery
   4. Data doesn't fit in available non-volatile memory
   5. Ability to reconstruct prior state (requires vast memory)

Memory hierarchies have been with us from the dawn of computing.  The 
classical version is:

   1. CPU registers
   2. Cache memory
   3. Main memory
   4. Disk

A more enlightened version:

   1. CPU registers
   2. Cache memory
   3. Main memory
   4. Other main memory in the cloud
   5. Archival disk somewhere in the cloud


-- 
Jim Starkey
President, NimbusDB, Inc.
978 526-1376