|
Home > Archive > Data Storage > May 2004 > Storage Recommendations for a Single Web Application
You are viewing an archived Text-only version of the thread.
To view this thread in it's original format and/or if you want to reply to
this thread please [click here]
| Author |
Storage Recommendations for a Single Web Application
|
|
|
| I've been reading posts from this group for the past few days and have
a decent understanding of the differences between DAS, NAS, and SAN.
However, it seems like the majority of the discussion revolves around
enterprise-wide file/data storage and sharing.
If I'm planning a single website dedicated to the online storage and
retrieval of documents (files, images, etc.) and would like to get
some recommendations from the group. Specifically, my storage
requirements will grow with my customer base - starting out at a few
GB and growing to 5-10 TB (if successful).
My question - what type of network architecture would provide the most
cost efficient scalability (i.e. I don't want to pay for 3TB until I
need it). Does SAN make sense for a single website? Additionally,
I'm going to rent dedicated servers from Rackspace and I know they
have NAS and SAN services, so I'm not considering building my own.
Please feel free to point out dumb questions - I'm fairly ignorant of
networking.
| |
| Bill Todd 2004-05-30, 11:11 am |
|
"Josh" <home@joshuajdmartin.com> wrote in message
news:ee71c6af.0405191302.1409211b@posting.google.com...
> I've been reading posts from this group for the past few days and have
> a decent understanding of the differences between DAS, NAS, and SAN.
> However, it seems like the majority of the discussion revolves around
> enterprise-wide file/data storage and sharing.
>
> If I'm planning a single website dedicated to the online storage and
> retrieval of documents (files, images, etc.) and would like to get
> some recommendations from the group. Specifically, my storage
> requirements will grow with my customer base - starting out at a few
> GB and growing to 5-10 TB (if successful).
>
> My question - what type of network architecture would provide the most
> cost efficient scalability (i.e. I don't want to pay for 3TB until I
> need it). Does SAN make sense for a single website? Additionally,
> I'm going to rent dedicated servers from Rackspace and I know they
> have NAS and SAN services, so I'm not considering building my own.
>
> Please feel free to point out dumb questions - I'm fairly ignorant of
> networking.
You say 'servers', suggesting that you believe that the access load will
exceed the capabilities of a single server. In that case, you've got to
decide how to distribute the load over multiple servers, whereas if you only
needed a single server, then the only question would be what kind of storage
to use with it, and the answer would probably be directly-attached storage
for the kind of capacities you're talking about - using attached RAID-type
boxes - unless your servers were located at a provider who could supply
unusually inexpensive (and suitably redundant) SAN storage (in which case
the issue of storage expandibility becomes their problem, not yours).
With multiple servers, the easiest though least flexible approach would be
to use DAS at each and partition the data statically across them (and the
least expensive as well, lacking the kind of SAN service I just mentioned).
This also provides at least one crude approach to expandability: just add
more servers and re-partition the data across them.
But if one of your servers went down, you'd then lose access to the portion
of your data that it held, unless you were using a shared SAN (again, likely
with redundant storage) and had made provisions for other servers to pick up
that portion of the data in a fail-over-style manner (which is an entire
subject in itself, e.g., since the data may not have been left in a
consistent state when the server failed). It's also possible to mirror data
from one server to a partner, using only DAS, such that the partner can take
over (again, modulo mechanisms to ensure data consistency) - but unless
you're running VMS or AIX (which on servers supplied by RackSpace seems
unlikely) you may need some third-party software to do this.
If you want all data to be accessible from all servers (to allow continued
operation in the event of a server failure), the easiest though most
brute-force approach would be to use provider-supplied NAS (with suitable
provider-supplied redundancy), though conceptually that's a bit silly since
you'd be serving files to your servers just so those servers could in turn
serve them elsewhere. Using the (suitably robust) NAS at least means that
the file system itself won't be left in an inconsistent state if one of your
servers fails, but that doesn't guarantee that your data will be internally
consistent (e.g., some multi-update transaction may have only partially
completed).
Or, you could run a 'SAN file system' on your servers that accessed shared
data on the SAN in a suitably-interlocked manner to ensure its coherency.
I'm not even going to begin to describe what that entails.
The bottom line is that unless you can live with an installation where
either only a single server is used or where data is statically distributed
across multiple servers, and in any event where the only mechanism for
taking over from a failed server is rebooting a new server to access the
failed server's data (presumably because it's on a SAN where the new server
can easily get to it, via normal file-system-recovery boot-up code that will
at least ensure the structural integrity of the file system on take-over,
though NAS could be used for this as well), you're likely biting off a far
more complex design than is implied by the simple "What kind of storage
should I use?" question you ask.
- bill
| |
|
| Bill - just wanted to say thanks for the great overview & recommendations!
|
|
|
|
|