cancel
Showing results for 
Search instead for 
Did you mean: 

Design 64TB MSDP Server

Scott_Chapman
Level 4

I'm looking for some design specs to help me size the hardware I need for a linux msdp server.  Looking to build a 64TB MSDP pool.

What I know:

1.  IBM 3650 (supports up to 8 or 16 internal drives)

2.  4 x 10Gethernet

3.  4 x 8G FC

4.  RedHat Linux

5.  SAN array to store the backup data

 

How do I figure out how much disk space I need for the fingerprint DB (metabase).  I'm not finding much useful information and I just don't think this should be all that big of a deal.  I'm being told that the fingerprint DB should live on fast disk (15K or SSD) and that it will be anywhere up to 10% of the size of the pool (so 6.4TB!!!).  That makes for a LOT of $$ on 15K or SSD drives...!!!

 

Any help would be very welcome.

 

Thanks, Scott

13 REPLIES 13

Douglas_A
Level 6
Partner Accredited Certified

So some basic considerations for MSDP..

Server should have 2x Quad Core CPUs 2.1Ghz or faster, for each 1TB of Dedupe storage you plan to have add 1GB of RAM.. So in your example you have a 64TB pool you would need at least 64GB + 12GB =76B RAm for the Media Server hosting the MSDP. (remember need memory for the media server processing as well) However i would recommend 96GB memory if you plan to go to 64TB, believe me NBU will use it. 

I would suggest something like this (If you have to split it up) follow the 3:1 rule.. 3 percent of disk for every 1 percent of MSDP storage. So in a 64TB environment roughly 20TB for the DataBase at max capacity (to be honest this is a little high but you can always start smaller and grow as required)

For performance Symantec recommends 150mb -200mb/read and write to the data storage.. SAS disk can more than handle this without issue for both the data and database volumes.

 

Hope this helps with your configuration

 

INT_RND
Level 6
Employee Accredited

From the dedupe guide:

"Table 2-3 MSDP server minimum requirements"

"For 64 TBs of storage, Intel x86-64 architecture
requires eight cores."

"RAM: 4 GBs for 4 TBs of storage up to 32 GBs for 64 TBs"

 

INT_RND
Level 6
Employee Accredited

There is no clear ratio between fingerprint database and storage used. It largely depends on dedupe rate which depends on the type of data you are storing. If you have a really good dedupe rate then you will have a higher fingerprints to data ratio. If you are getting a bad dedupe rate then you will find that you are using less fingerprint space.

Similar data will dedupe really well. Random data does not dedupe well. Databases are a great example of something that doesn't dedupe well. OS files are a great example of somthing that dedupes very well.

The deduplication storage guide will answer a lot of your questions. Chapters 2 and 3 will help you plan your set-up.

From Page 50:

"About MSDP storage capacity
The maximum deduplication storage capacity is 64 TBs. NetBackup reserves 4
percent of the storage space for the deduplication database and transaction logs.
Therefore, a storage full condition is triggered at a 96 percent threshold."

Deduplication storage guide:

http://www.symantec.com/business/support/index?page=content&id=DOC6466

Scott_Chapman
Level 4

So it sounds like I'm hearing that I don't need a special pool of disk for the fingerprint db?  

Am I better to have the fingerprint db local to the server rather than out on SAN disk where the actual data will live?  The server I will be using can have up to 16 disks installed internal.  Can I assume there will be some configuration to allow me to configure where the fingerprint db will live vs where the data volume will live?

Thanks so much everyone for responding!

Scott

V4
Level 6
Partner Accredited

hope OP had got attention on NBU 7602 where MSDP has been scaled beyond to 64TB limit.. So you got more room for designing MSDP now and to cover more clients with front end TB :)

And with 7.6 dedupe database was rewritten into flatfile i believe.. this is much faster than 7.5 .... 

INT_RND
Level 6
Employee Accredited

Normally you wouldn't but in this case you do. By default the installation will put all the storage paths on the same volume. For your case you should add a custom path for your database:

"Depending on the size of your deduplication storage, Symantec also recommends a separate path
for the deduplication database."

"CAUTION: You cannot change the paths after NetBackup configures the deduplication
storage server. Therefore, carefully decide during the planning phase where and
how you want the deduplicated backup data stored."

"The maximum deduplication storage capacity is 64 TBs. NetBackup reserves 4
percent of the storage space for the deduplication database and transaction logs.
Therefore, a storage full condition is triggered at a 96 percent threshold.
For performance optimization, Symantec recommends that you use a separate
disk, volume, partition, or spindle for the deduplication database. If you use separate
storage for the deduplication database, NetBackup still uses the 96 percent threshold
to protect the data storage from any possible overload."

If we say that you need %4 of the total for your database 64TB * %4 = 2.56TB

"64 TBs of storage:
250 MB/sec.
Symantec recommends that you store the data and the deduplication database on separate
disk, each with 250 MB/sec read or write speed."

You should overprovision that volume. It's also critical so you need a RAID array. Plus the throughput needs to be around 250MB/s. You could achieve this by striping mechanical drives with a RAID 0+1 setup or by putting some fast SSD in a RAID6. There are pros and cons to both of these setups and both are very expensive.

 

INT_RND
Level 6
Employee Accredited

All that being said. You could make one big SAN with 70TB of storage and cut out 2 volumes from it. The 64TB volume for your data and the rest for the database volume. You would be using multiple "disk, volume, partition, or spindle" as the guide states. The only problem would be link saturation. Competition for I/O. Using the internal storage of the box would give you multiple I/O paths to the arrays. You might get better performace with 2 sets of disk. It's all about cost at this point.

Nicolai
Moderator
Moderator
Partner    VIP   

Remember is very easy to extend file systems on the go using Linux/UNIX systems.

Just add storage if a possible fingerprint volume run low on space.

Marianne
Level 6
Partner    VIP    Accredited Certified

Design headache should be sufficient motivation for a NetBackup Appliance.

One supplier provides ALL of the hardware (server, NICs HBAs, arrays, etc), the OS, the volume management software, the monitoring, the security software, etc... etc..
All pre-configured for performance.

So, for a DIY appliance, I would look at the 5230 Appliance datasheet and build something with similar specs.

See attached doc.

RonCaplinger
Level 6

Do you have a link showing where MSDPs can go beyond 64TB?  I've looked through the release notes and documentation for both 7.6 and 7.6.0.2, and nothing indicates the 64TB limit has been changed.

Marianne
Level 6
Partner    VIP    Accredited Certified

RonCaplinger
Level 6

Thanks, Marianne! 

So the NBU 5320 appliances get the capacity increase right now, not the MSDPs that are created on other hardware, until some later date.  Also, this means that 7.6.0.2 does not include this increase in MSDP size limit; possibly in 7.6.0.3....hopefully.

smourylev
Not applicable

Hi Scott, 

Did you made some progress regarding this challenge?

We are also going this way, design and technical challenges are allways interesting.

Cheers,