Networker

Re: [Networker] e4500 Max I/O capacity issue as Legato Storage Node

2003-04-07 07:34:58
Subject: Re: [Networker] e4500 Max I/O capacity issue as Legato Storage Node
From: Howard Martin <howard.martin AT EDS DOT COM>
To: NETWORKER AT LISTMAIL.TEMPLE DOT EDU
Date: Mon, 7 Apr 2003 07:35:04 -0400
>From our experience a 400MHz 450 (4 cpu, lots of RAM) will only take 45-60
MB/s from a gigabit etherenet, I don't think they will aggregate well as
the tcp/ip stack runs on one processor and the processor speed limits it.
Using a 900MHz V880 however we get a throughput to disk of 95-06 MB/s and
hence that amount thro a gigabit ethernet, this system will also keep 8 (no
more drives to test with) IBM LTO's streaming and 21-22MB/s (168-176 MB/s
aggregate).

On Fri, 4 Apr 2003 14:43:25 -0600, Reed, Ted G II [CC]
<ted.reed AT MAIL.SPRINT DOT COM> wrote:

>I am trying to determine the max I/O capacity of a Sun E4500 for usage as a
>Legato storage node.  Here are the current stats and 'designed' I/O plans:
>
>Sun E4500, 8x 400Mhz 4-8G Ram
>        4 System Boards, 3 I/O boards
>4x Gb Ethernet (Trunked as single IP)
>3x 2Gb HBA fibre to Tape devices
>6x STK 9940B in STK 9310 silo***
>
>Since you lose ethernet bandwidth due to TCP/IP overhead, you can
anticipate
>~80MB/sec* per NIC (total ~320MB/sec).  The HBA fibre can handle
>~250MB/sec** (total ~750MB/sec) and the 9940B drives running compressed at
>60MB/sec*** per drive (total ~360MB/sec).  This should mean I will be
>passing 320MB/sec over the e4500 I/O boards.  I am still determining if
this
>needs to be done GigE+HBA per I/O board or an all GigE board, all HBA
board,
>and a 'rollover' I/O board.
>So I guess I am really wondering if anyone is pushing anything close to
this
>level of bulk data over a single storage node, regardless of the hardware
>types being used.  Thank you all in advance for any information you may be
>able to provide.
>--Ted Reed, Engineering Storage Services
>
>
>* 10BaseT ~0.8KB/sec, 100BaseT ~8MB/sec, 1000BaseT ~80MB/sec due to ~20-25%
>TCP/IP overhead
>** Little to no overhead on fibre for protocols.  1Gb=125MB, 2Gb=250MB
>*** 30MB/sec native, 60MB/sec compressed, 90MB/sec burst.  200G native
>capacity, 300-600G compressed
>
>--
>Note: To sign off this list, send a "signoff networker" command via email
>to listserv AT listmail.temple DOT edu or visit the list's Web site at
>http://listmail.temple.edu/archives/networker.html where you can
>also view and post messages to the list.
>=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=

--
Note: To sign off this list, send a "signoff networker" command via email
to listserv AT listmail.temple DOT edu or visit the list's Web site at
http://listmail.temple.edu/archives/networker.html where you can
also view and post messages to the list.
=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=

<Prev in Thread] Current Thread [Next in Thread>