Regardless of a specific application and configuration, I am looking for an architecture that will support nas. Setting all components equal, comparing apples to apples, you end up with either going through a server for each operation or accessing a disk directly and an architecture that lets you access the disk directly must be faster. If you could have fast disk channels, getting information from that disk must be faster than getting the information through the server. It is not completely true for all applications. You still need the server for co-ordination and then it is a trade off. But I believe that each storage vendor will need this capability to be competitive. Some applications that would need this capability are publisher consumers messaging model, backup, source control.
-----Original Message----- From: Josh McCormick [mailto:jmccorm@mail.wcg.net] Sent: Thursday, March 23, 2000 9:02 AM To: mkeller@mail.wcg.net Cc: toasters@mathworks.com Subject: Re: nas storage
Necessary to yell? I use what I received. I am required to stick with
it for a while yet.
You make a blanket statement without supporting
facts.
Here's what I'm wondering, Michael. When they are comparing the Netapp Filers to local storage, what exactly are they comparing?
* Are they comparing 100mbit NFS delivery to a SCSI variant? * Are they comparing their raid and caching to a(n anonymous) local disk solution? JBOD or something else? * Are they comparing their NFS filesystem implementation to UFS and/or VxFS?
Are the comparing the whole solution against something when it is claimed to be on par with local disks, and if so, what solutions are they comparing against?
Of course, it would be a surprise is the Netapp Filers were on par with local storage in all cases. What I want to know is, what are the edge cases? What are the filers particually better at, and what are they particularly worse at?
__________________________________________________ Do You Yahoo!? Talk to your friends online with Yahoo! Messenger. http://im.yahoo.com
I heard Yael Hellmann (Thu, Mar 23, 2000 at 08:05:35AM -0800) say
Regardless of a specific application and configuration, I am looking for an architecture that will support nas. Setting all components equal, comparing apples to apples, you end up with either going through a server for each operation or accessing a disk directly and an architecture that lets you access the disk directly must be faster. If you could have fast disk channels, getting information from that disk must be faster than getting the information through the server. It is not completely true for all applications. You still need the server for co-ordination and then it is a trade off. But I believe that each storage vendor will need this capability to be competitive. Some applications that would need this capability are publisher consumers messaging model, backup, source control.
As has been said before, network speeds are coming close to disk connection speeds, but there's a couple of other things.
Firstly, if you write across the network to a NAS device, for pretty much any vendor, you're not writing to disk you're writing to NVRAM. You're going to get your write confirmation back a lot faster on larger writes to NVRAM than you are to disk. Write 5MB - a good disk does that in around half a second. How about that into NVRAM?
Your free block / file exists / etc decisions are performed at one end or other, it doesn't much matter which, although a specialist NAS device may well have its entire blockmap in memory - does your server?
And lastly, which is cheaper? Outfitting 10 servers with local disks of a high enough standard to beat NAS, or the NAS device and networking?
For a few servers I'm sure the balance points to local-attached, with all the attendant hardware-RAID controllers and top of the line kit.
Now try and sell your boss on outfitting a hundred servers.
While that's a cheapshot in a discussion on performance, performance you can't afford may as well not exist.
We have applications and servers where we want to spend the money on local disks, and worry about backups and contingency some other time, but as the number of spindles on a NAS device is consistently going to be more than can easily be attached locally, to keep pace with better local-disk technology, I'm upgrading my network, not upgrading my data *to new disks*.
Can you say downtime?
So keep upgradeability in mind with comparison of NAS and local, as well as cost and convenience. Raw disk performance isn't everything.
J
--------------------------------------------------------------------------- # John Denholm johnd@theplanet.net # # Webcache & Filer Administrator, Planet Online +44 113 207 6357 # ---------------------------------------------------------------------------
Error 404: There is no spoon