Media Server - Drive connectivity

[iaw]

Hi All,
I have 2 media servers with 2 HBA cards (=4 ports), 2 san directors, 24 drives and 24 SSO license.
Please suggest whats the best scenario for media-drive zone connectivity to get good speed performance.
1. Each media connect to 8 drives (2 drives per hba ports)
2. All media connect to all 24 drives (6 drives per hba ports)
or any better scenario?

Thanks in advance,
Ifan

[revarooo]

2 Media servers and 24 drives. Personally. I would connect 12 drives to each media server and not bother with sharing them to each other.

[Marianne]

My thoughts : too many drives, not enough media servers. A single media server will even battle to keep 8 tape drives streaming at more than 100 Mbytes/sec.

[Nicolai]

I am also in the crowd of too many drives per media server. You need to look at the incoming data rate and the outgoing rate. A 10GB Ethernet can effective provide 600MB/sec,  but a LTO5 does 140MB/sec native, add compression and it does 210MB/sec. So a 10GB Ethernet connection can drive a best 3 LTO5 drives.

My advice:

I would have no more than 4 drives per media server. Each drives on its own HBA port.

For SSO i recommend what I call SSO pairs. Two media servers sharing e.g 8 drives (4 active each), then another media server pair sharing the next 8 etc etc.

[iaw]

Sorry my mistake, this will be 3 media servers and 24 drivers.
Should I connect phisically 1 media to 8 drives and use 24 SSO licenses.
Will really SSO help to speed up backup window?

[iaw]

Hi Nicolai,

It runs on 10GB Eth and LT06.
Can you please more elaborate, about SSO pair for 3 media server + 24 drives?
Can SSO help to speed up backup window?

Thanks
 

[sdo]

SSO does not speed up backups.  SSO enables a form of HA (High Availability) of a 'set of tape drives' across multiple media servers.

SSO would let you share more tape drives per media server.

However, the reason for the reticence from us re not worrying too much about implementing SSO, is that it is very very difficult for one media server to push eight LTO6 tape drives, so therefore there is probably not much point in implementing SSO to enable even more tape drives to be used across any one media server.  For example, it may actually transpire that your media servers are so powerful that they can drive (just about) eight LTO6 drives - but if you were to implement SSO, then it is possible that you could present 16 tape drives to a media server - but this may cause all tape drives to all run much slower (because of shoe shining) and so what could actually happen is that backups to 16 tape drives actually runs slower - in terms of total aggragated bandwidth - than backups to 8 tape drives.

.

Some questions which will help us be more specific:

1) Exact brand, model and generation of media servers?

2) Speed of link between media server HBA port and SAN switch port?

3) Speed of link between SAN switch port and tape drive?

4) Exact brand and model of SAN switch?

5) Will the server port and tape drive that are to be zoned together be cabled to the same SAN switch?  Or will they be on different SAN witches within the same SAN fabric?

6) Exactly which model of SAN switch?  If DCX and using blades, then exactly which 'blade' model?

7) Will you be backing-up to the tape drives, or duplicating to them?

.

With the above answered clearly and in detail - there are some things that we may be able suggest, and or describe/outline in more detail to help you better understand the limitations of your environment.

[iaw]

1) Exact brand, model and generation of media servers? BL620G7, 2Proc, 2.1Ghz, 16core, RAM 64, Linux 64bit

2) Speed of link between media server HBA port and SAN switch port? 8GB

3) Speed of link between SAN switch port and tape drive? 8GB

4) Exact brand and model of SAN switch? DCX

5) Will the server port and tape drive that are to be zoned together be cabled to the same SAN switch?  Or will they be on different SAN witches within the same SAN fabric? zone in the same SAN Switch

6) Exactly which model of SAN switch?  If DCX and using blades, then exactly which 'blade' model? info above

7) Will you be backing-up to the tape drives, or duplicating to them? tape drive, no duplicating

So at this point (3 media + 24 drives) , no need SSO? (considering backup window time)

Thanks

[sdo]

5) If the server HBA initiators and tape drive targets are to be cabled on the same SAN switch - then IF the SAN switch blade ports are not non-blocking, i.e. if the SAN switch ports are 'contended' in port-groups - AND/OR - the blade has multiple ASIC then you may want to pay careful consideration to exactly which ports the initiator and the target are cabled to.  There are possibly some really neat tricks you can leverage - but you need to supply the link requested in question 6) continued below.

6) Need model of blades in the DCX.  Please - to save me time, can you post a link the tech specs pdf of the blade that you will be using.

8) That's the first time you have mentioned backup window.  What is the backup window of the full backup?

9) Do you know how much data there is to be backed-up in full?

10) Ask your network admin whether the 10Gb LAN switch ports that will be supplied for the media server, whether they will be residing in 'uncontended port groups'?  Or will each 10Gb link reside with other 10Gb links in the same port group on the LAN switch(es)?

.

I need to be firm now.  You missed some detail in your responses to the initial set of questions.  You cannot afford to be scant with detail - if you want real advice.  You need to answer these questions in detail.  Real detail!

I need to make this clear too.  Your opening question is so wide, so vast, that it will take several iterations of us asking questions before any thing even approaching any form of soft advice will be realised.  You have to help us to help you.  There is no quick answer to your design question.  However, if you continue to answer questions clearly and accurately, then together we may be able to improve your understanding of the potential complications that you need to consider.  HTH.

[revarooo]

Definitely agree with you there Marianne.

[Marianne]

More tape drives or SSO will not help fix backup window issue. 

Back to Nicolai's post above: 

You need to look at the incoming data rate and the outgoing rate. A 10GB Ethernet can effective provide 600MB/sec,  but a LTO5 does 140MB/sec native, add compression and it does 210MB/sec. So a 10GB Ethernet connection can drive a best 3 LTO5 drives.

If you can only get data at 600 MB/sec to each media server, more than 3 or 4 drives per media server will not help.

Then there is the matter of clients being able to read from local disk, generate backup stream and send data across the network.
If you cannot get clients to send data fast enough to ensure  600 MB/sec ingest rate on each media server, you may not even see tape drives writing at close to 100 MB/sec.

Was a proper planning and sizing exercize conducted before hardware purchase was decided?

[iaw]

5)SAN Switch ports are not non-blocking

6)DCX HP SN8000B
http://h20564.www2.hp.com/hpsc/doc/public/display?docId=emr_na-c03089914

8) 8 hours backup window

9) Total 30TB daily in a full, 5TB Incr.
I have important 4TB data daily full, from 1 server that need to be backup in 8 hrs.
Total server = 70

10) It has dedicated Lan Switches for backup network only, each 10Gb link FO reside in the same port group

Im trying to give all the require details info, kindly revert back if u need more.
Really need suggest on this.

Thanks in advance

[Nicolai]

The hardware you have will not be able to get all 24 tape drives running with 140MB/sec each.

I actual think you wil get better speed by reducing the amount of tape drives used concurrent.

Please consider a re-design with bandwith in mind.

[sdo · Accepted solution]

Here's some very basic calculations - but these are assuming an awful lot - I'm not thre with you - I cannot vouch for these calculations - I'm only including them here to give you a few pointers:

size of full backup	30.0	TB
size of full backup	30,720.0	GB
size of full backup	31,457,280.0	MB
full backup window	8.0	hours
full backup window	28,800.0	seconds
required aggregated bandwidth	1,092.3	MB/s
number of media servers	3.0	servers
minimum bandwidth required per media server	364.1	MB/s
minimum raw LAN ingest required	2,912.7	Mb/s
minimum raw LAN ingest required	2.9	Gb/s
typical efficiency of LAN ports	85%	efficient
minimum raw LAN ingest required	3.4	Gb/s
theoretical max raw speed of LTO6	160.0	MB/s
typical compression achieved by LTO6	1.4	ratio
theoretical max raw speed of LTO6 with compression	224.0	MB/s
typical efficiency of LTO6	90%	efficient
theoretical max raw speed of LTO6 with compression	201.6	MB/s
number of LTO6 tape drives required per media server	1.8	LTO6 tape drives
roundup number of tape drives required per media server	2.0	LTO6 tape drives
total number of tape drives required	6.0	LTO6 tape drives
plus one spare drive to cater for failure and restores	7.0	LTO6 tape drives

[Nicolai]

Excel sheet ?

[sdo]

Attached.

[sdo]

@law - the calculations that I have posted are all theoretical maximums - pie in the sky, so to speak - and do not take into consideration any actual real world figures.

You need to pay very careful attention to what everyone else has said - all of them.

[areznik]

Keep in mind something - 

If your clients have 1gig ports (or lower) or there is a network bottleneck between the clients and the media servers, it doesnt really matter how fast you can theoretically push your tape drives. They will just sit there mostly idle waiting for data to get to the media server. So check on that as well. 

[sdo]

5) When you say "not non-blocking"... do you mean that you know for certain that "port contention" is in existence?

6) That's the switch chassis model.  I need links to the tech specs of the blades within the chassis... if you would like comment on how to further avoid bottlenecks at the SAN transport layer.

10a) What is "FO"?

10b) Remember a "dedicated" LAN switch is not much use if the 10Gb bandwidth is not "dedicated" at the port level.  I mean, if you share four x 10Gb LAN links in one "LAN switch port group of four ports" where this "LAN switch port group" as a whole is only capable of 10Gb - then you can never ever be guaranteed of unfettered 10Gb speed for any port cabled in that port group.  I would ask myself, what's the point of having a dedicated LAN switch if the LAN port itself does not reside within it's own decicated uncontended port group?  Put it this way... a backup server is NOT a typical server.  I totally get putting four application servers each having one 10Gb link in a contended "4 x port group" - but backup servers move huge amounts of data, so why on earth would anyone put a backup server's main data ingest port in a contended port group.

[sdo]

@law - I've asked for futher detail above, which if you supply details for, then I will add further comment.

11) Apart from that, is there anything else that you would like to ask?

.

The sheet I posted above has 21 rows in it, and one column (of numbers).  IMO, a semi-decent sizing / projection for NetBackup woud typically be expected to have a minimum of 150+ rows, and an additional sizing column for each projected year in to the future.  If your sizing spreadsheet model does not have 150+ rows then you are not asking yourself enough good questions.  If your sizing spreadheet does not have more than one 'numbers' column then you cannot possibly be thinking and planning ahead.  Column 'B' should be sized for "today" (i.e. starting at day one), column 'C' could be +1 year, column 'D' could be +2 years... etc.