|
Home > Archive > Data Storage > June 2006 > SAN/NAS Recommendation
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 |
SAN/NAS Recommendation
|
|
|
| Hi Everyone -
We are beginning our quest to upgrade our existing Dell PowerVault DAS
for our Linux database servers and I'd like to get some recommendations
regarding which direction to take.
We anticipate our maximum size threshold for the Oracle database being
in the 10-20TB range. Our application is very write intensive. To
accomplish these writes, our custom ETL process performs a significant
amount of reads on primarily recent data to determine it's inserts. We
also have a web-presentation layer but our user base is very limited
(our real-time ETL process is a factor of 100 greater load wise).
>From everything I have read NAS solutions offer a more inexpensive
approach allowing one to scale the size of your storage but do not
measure up to FC SAN solutions when it comes to performance. We also
have to keep an eye on cost - which obviously SANs are much more
expensive.
Would be curous:
1) What type of vendors (EMC/Legato, HTC, NetApp, etc.) and solutions
(NAS, SAN, etc.) would you recommend.
2) NAS or SAN - between the two, is there a general rule as to the cost
per TB? Would be curious just order of magnitude. Does not have to be
exact by any means.
3) What type of redundancy capabilities would we gain going with a SAN
over NAS or even over our existing DAS?
4) Our current DAS is in a Raid 5 configuration - we did this to
maximize space but still give us some amount of redundancy. With the
new solution we are looking to Raid 10 or 0+1 - I would think 0+1
(mirror then stripe) would be best so that if a drive goes out, you
have a hot-spare on that side which becomes the new mirror and don't
have to re-sync the entire other striped set of drives (as you would in
Raid 10).
5) Has anyone implemented a hierarchical storage solution - what are
your throughts on it if so? Dell when pitching us EMC/Legato made that
seem like the latest rage - putting your latest data (to include our
massive inserts) on the fastest storage then having Legato
automatically move it to slower and cheaper storage.
Thanks for any insight you can provide...
David
| |
| Faeandar 2006-05-25, 1:12 am |
| On 24 May 2006 13:54:16 -0700, "David" <dlejeune@belmontcc.com> wrote:
>We anticipate our maximum size threshold for the Oracle database being
>in the 10-20TB range. Our application is very write intensive. To
>accomplish these writes, our custom ETL process performs a significant
>amount of reads on primarily recent data to determine it's inserts. We
>also have a web-presentation layer but our user base is very limited
>(our real-time ETL process is a factor of 100 greater load wise).
Question everything, especially what your dba's tell you. As per
below, if your current environment has been serviceable, and it's raid
5, then your application is not write intensive.
>
>approach allowing one to scale the size of your storage but do not
>measure up to FC SAN solutions when it comes to performance. We also
>have to keep an eye on cost - which obviously SANs are much more
>expensive.
Again, question everything. There is nothing inherintly slower about
NAS v. SAN. It's all in the architecture.
Also, SAN's are not obviously more expensive. Requirements drive cost
through features. Some features cost more in one than the other.
>
>Would be curous:
>
>1) What type of vendors (EMC/Legato, HTC, NetApp, etc.) and solutions
>(NAS, SAN, etc.) would you recommend.
Lets give this a generic go.
Oracle over NFS on NetApp. I can't tell how many spindles are in your
current environment but I would suspect that a 980c with 4 shelves
should be sufficient.
Add snapmirror for DR, and snaprestore for backup/restore and
deve/test environments. Flexcache comes built into the OS, as do
flexvols and a few other nifty db features.
>
>2) NAS or SAN - between the two, is there a general rule as to the cost
>per TB? Would be curious just order of magnitude. Does not have to be
>exact by any means.
Cost is almost always based on vendor desire to get/keep you as a
customer. There are street prices, discount prices, and tier1 prices.
If you have no NetApp on the floor, even more so if you are replacing
a "competitor", then the vendor is usually ready to go for discount+.
>
>3) What type of redundancy capabilities would we gain going with a SAN
>over NAS or even over our existing DAS?
Potentially none, if you're DAS was setup with HA in mind. There is
also no increased redundancy between NAS or SAN, both are equally
redundant if architected correctly.
>
>4) Our current DAS is in a Raid 5 configuration - we did this to
>maximize space but still give us some amount of redundancy. With the
>new solution we are looking to Raid 10 or 0+1 - I would think 0+1
>(mirror then stripe) would be best so that if a drive goes out, you
>have a hot-spare on that side which becomes the new mirror and don't
>have to re-sync the entire other striped set of drives (as you would in
>Raid 10).
Most people can put raid type aside and look for performance and
reliability. Absolute reliability is optimized in raid 10 (striped
mirror as opposed to a mirrored stripe) but, again, if raid 5 has been
serviceable then anything is an improvement.
Raid DP in NetApp world (and most vendors have a similar if not
identical concept) is extremely reliable.
>
>5) Has anyone implemented a hierarchical storage solution - what are
>your throughts on it if so? Dell when pitching us EMC/Legato made that
>seem like the latest rage - putting your latest data (to include our
>massive inserts) on the fastest storage then having Legato
>automatically move it to slower and cheaper storage.
The product I've heard the best of is Princeton Softech. They have a
db archiver that has been around a long time now. I know nothing
about the Legato solution though.
HSM is the latest rage if you go for the latest rage. ILM
specifically was last years golden child, this years it's
virtualization....again.
Only worry about it if it solves a business driver. Never do anything
for the sake of doing it. At least not when you're the one getting
paged when the thing barfs. KISS (keep it simple stupid) will apply
to the end of time.
~F
| |
| Faeandar 2006-05-29, 5:04 pm |
| On Thu, 25 May 2006 10:06:33 +0200, Zak <jute@zak.invalid> wrote:
>Faeandar wrote:
>
>
>A San is more expensive if you have a large number of low-end servers.
>Switch pots and HBA ports are more expensive for a SAN than for a NAS.
>Apart from that there won't be much difference.
SAN is not a protocol (ie FC) it is an access type. iSCSI SAN's use
the same infrastructure that NAS uses, namely ethernet and network
switches.
And ethernet infrastructure isn't always cheaper. Look at 10gb prices
and FC will look cheap by comparison.
>
>Mind you that this can be an enabler: the low cost allows much larger
>dev farms made up out of desktop-class machines, for example - if you
>are so inclined.
>
>
>And for redundancy/HA make it a cluster.
That's the "c" portion in 980c. c=cluster. You are correct though in
that a 3050 or even a 3020 may well serve the OP's needs. It doesn't
seem like performance is as much of a requirement from his description
of the current environment.
>
>Cost per GB will be similar. Oh yes: do not forget service contracts. I
>remember EMC offering us drives for next to nothing, but support on the
>drives was more than purchase. That's different now I suppose.
Ouch! I personally will always avoid doing business with EMC but
that's, well, personal. But the moral of the story is still
applicable, service...costs.
>
>You can also look at things like oracle replication, which guard against
>a wider range of features and replicate off-site as well.
The OP should also talk to Oracle. I know they have petabytes of
NetApp storage and are using almost all the Snap* tools. The company
I currently work for is looking to do the same thing, and the company
I worked at previously did also. There's a pattern there somewhere,
not sure if it's just me though....
>
>
>Doesn't seem so interesting for a database - or does it. Depends on the
>data volume though. It will add complexity, and complexity.. well...
>rather be without it.
I looked at the post again and a 10-12TB db is pretty damn huge. HSM
of some sort may well be worthwhile, if for no other reason than full
backups...
~F
| |
|
| In article <1148504056.768950.304010@u72g2000cwu.googlegroups.com>,
dlejeune@belmontcc.com says...
>
>Hi Everyone -
>
>We are beginning our quest to upgrade our existing Dell PowerVault DAS
>for our Linux database servers and I'd like to get some recommendations
>regarding which direction to take.
>
>We anticipate our maximum size threshold for the Oracle database being
>in the 10-20TB range. Our application is very write intensive. To
>accomplish these writes, our custom ETL process performs a significant
>amount of reads on primarily recent data to determine it's inserts. We
>also have a web-presentation layer but our user base is very limited
>(our real-time ETL process is a factor of 100 greater load wise).
>
>approach allowing one to scale the size of your storage but do not
>measure up to FC SAN solutions when it comes to performance. We also
>have to keep an eye on cost - which obviously SANs are much more
>expensive.
>
>Would be curous:
>
>1) What type of vendors (EMC/Legato, HTC, NetApp, etc.) and solutions
>(NAS, SAN, etc.) would you recommend.
>
>2) NAS or SAN - between the two, is there a general rule as to the cost
>per TB? Would be curious just order of magnitude. Does not have to be
>exact by any means.
>
>3) What type of redundancy capabilities would we gain going with a SAN
>over NAS or even over our existing DAS?
>
>4) Our current DAS is in a Raid 5 configuration - we did this to
>maximize space but still give us some amount of redundancy. With the
>new solution we are looking to Raid 10 or 0+1 - I would think 0+1
>(mirror then stripe) would be best so that if a drive goes out, you
>have a hot-spare on that side which becomes the new mirror and don't
>have to re-sync the entire other striped set of drives (as you would in
>Raid 10).
>
>5) Has anyone implemented a hierarchical storage solution - what are
>your throughts on it if so? Dell when pitching us EMC/Legato made that
>seem like the latest rage - putting your latest data (to include our
>massive inserts) on the fastest storage then having Legato
>automatically move it to slower and cheaper storage.
>
the bottom line thing to know about the difference between NAS & SAN is
a. a NAS is a "disguised" file server appearing as a storage subsystem
b. a SAN storage device, like a RAID, is storage that can be used in
place of DAS for applications which is why you're probably looking
at SAN for an Oracle requirement
* iSCSI & FC storage are both SAN storage and function the same way except
for the fact that the iSCSI storage runs over your (typically) existing
IP infrastructure (network) and you (typically) need to buy a Fibre
Channel network (switches & HBAs) to run a Fibre Channel SAN
_____ . .
' \\ . . |>>
O// . . |
\_\ . . |
| | . . . |
/ | . www.EvenEnterprises.com . . . |
/ .| info@EvenEnterprises.com . . . |
/ . | 310-544-9439 / 310-544-9309 fax . . . o
----------------------------------------------------------------------
Authorized - DIRECT VAR/VAD/Distributor for new mid-high end storage
iSCSI/NAS/SAN/RAID from EMC, HP, Equallogic, Quantum, OverLand Storage
|
|
|
|
|