2.5 10K SAS vs 3.5 15K SAS...

[bob4432]

been out of the tech loop for a bit and was wondering how the 2.5 10K SAS drives compare to the 15K 3.5 SAS drives - w/ the same type of random data useage pattern on the same controller? does the 2.5 make up for its slower rotational speed by being smaller? or is the 15k still king for random seek times? thoroughput is not a concern as they are already over what is needed for the application.

was spec'ing out a server for a friend and noticed the server only has 2.5" hotswap bays....no 3.5

this is the machine we were looking at - http://www.cdw.com/shop/products/def...OD600426)#TS

was going to put in a raid1 setup, 300 or 600GB, then of course nightly imaging. will be running server2k8 w/ sql2k8 for a ~1GB db for 15users on at any single time.

am personally still running u320 15k scsi, haven't upgraded yet so haven't been able to test out the new sas hdds.

any info would be greatly appreciated,
bob

 

[LokutusofBorg]

They make 2.5" 15k SAS drives...

 

[bob4432]

They make 2.5" 15k SAS drives...

15k 2.5 r only 146GB, at least what i saw - are there 2.5 15k 300GB? need 300GB and would prefer raid1

 

[Zstream]

What are you using it for?

 

[LokutusofBorg]

You've not shared all the requirements with us if you can't fit Win2k8 + SQL + 1GB DB on a 146GB drive...

At 2.5" the 15k drives are well above a 10k drive. Bumping up to 3.5" you lose some of that due to longer seek times, but they're still going to be better. Having said that, you're better off going with multiple spindles of 10k drives in RAID 10 instead of trying to get the bestest 15k drives. A 15 user (max) DB isn't going to see much of the benefit of 15k drives.

 

[bob4432]

man, i skipped over a big part, machine will also be a file sharing machine for the biz. there are other files on the machine too that don't get accessed as often, so that is the need for the extra space and would like to have it all on a single array. there are many files that are small (less than 100KB) that are looked every once in a while and just want the snappiness of the higher rpm hdd as there may be a couple hundred to a thousand in certain areas.

we swapped out the hdds on another file server that all files are are less than 200KB, but again, thousands of them, and went from an old 7.2k pata hdd to a fujitsu 15k mas drive and the difference when looking at that share is night and day, so wanted to get the same benefit from the main server upgrade. the last upgrade was 7-8yrs ago, so the current machine has definitely served well , but time to upgrade.

after looking at the website, i did find the same machine w/ 3.5" space, so will probably go that route and do the 15k 3.5" hdds - wasn't sure if 2.5" took over 3.5" like sas has w/ "old scsi".

 

[sub.mesa]

Use 5400rpm HDD if you need sequential performance on your volume.
Use SSDs instead of HDDs if you need random IOps performance on your volume; HDDs no matter what configuration will always suck in this kind of workload.

A combination of both sounds most logical; i can't see a reason to justify the costs of a high-end HDD if it still sucks for the purpose you're using it for.

Personally i think money spent on 10/15k HDDs these days is wasted, especially if "SSD+5400rpm" was a good alternative. Think what you want to achieve, and whether you really want to spend alot of money on mechanical storage, in 2010. Mechanical storage aimed at capacity-per-dollar is fine, but mechanical storage aimed at performance doesn't make much sense to me with SSDs floating around. A small SSD + larger HDD can give you all the IOps + throughput you need i think.

 

[LokutusofBorg]

5400 RPM is anathema in the server space...

@bob You shouldn't discount an array of a few 10k drives if you're at all price conscious. Modern 10k drives are really fast, and the usage profile you've outlined is not even close to pushing the limits of a couple 10k spindles in a RAID1 or 3 of them in a RAID5 or whatever. We are running production, high-usage databases on (modern) 10k arrays and they do fine.

 

[bob4432]

after doing a bit of research, i think before we go w/ the buy, we are going to change the db from access to sql2k5express and see if that does anything, since the access db errors are happening quite often and it is one of the main reasons for the upgrade. if this speeds up the program w/ the db and fixes errors i have a u320 lsi pci-x raid card i can give him and put some hdds on that and also add more ram to the machine, since we have that already. the current cpu isn't being taxed at all w/ the software as it is and when you look at the price w/ the addition of hdds, os and sql workgroup license/cals, you are getting up there $$ wise when this change might help out, it is atleast imho worth a shot. the machine only uses 400-500MB of ram @ peak currently and isn't using more than 20%cpu on average, actually quite a bit less, but the age does concern me.

anybody know how much more efficient sql2k5 is compared to access2k or possibly 97, not sure what version access they are running....

 

[Dark Shroud]

Is there any reason you can't build your own server with hot swap trays?

Build minus case:
http://www.newegg.com/Product/ComboD...t=Combo.541727
http://www.newegg.com/Product/Produc...82E16817151094
http://www.newegg.com/Product/Produc...82E16820231307
http://www.newegg.com/Product/Produc...82E16814131339

Trays:
http://www.newegg.com/Product/Produc...82E16817990011
http://www.newegg.com/Product/Produc...82E16817990012
http://www.newegg.com/Product/Produc...82E16817990013

Three 10k drives in a Raid5 should do the job you're looking for just fine.

 

[Emulex]

i have a dl380g5/7

8 x 10K 146GB 2.5" in raid 10 (esxi) is far slower than 6 x 300gb 15K SAS 3.5" raid 10

I'd suggest you think long and hard because you really do want a raid-1 for log (2 disks) for linear write) and raid-1 for tmp (random). Then i'd do raid-10 for core storage since the cost is very little over the raid-5 and i've had TWO double disk failures in my life - which seems impossible.


Sql server is an OS (SQLos) it can be tuned efficiently but in most cases it can be just as slow as ms-access if not run correctly.

separate SQLos from any other roles (esp AD); no av; backup. try not to VM unless you have a san where you can do iscsi or RDM to volumes so keep the main db, log,tmp in many separate files. If you have 4 processors it can dispatch 4 threads so at least 4 files.

 

[LokutusofBorg]

SQL Server is a server application. Access is a client application. Putting your Access files on a server does not turn that server into a database server; it is nothing more than a file server. SQL Server Express is trimmed down, but it is an actual database engine (server application). Access is for people that don't know better.

 

[bob4432]

SQL Server is a server application. Access is a client application. Putting your Access files on a server does not turn that server into a database server; it is nothing more than a file server. SQL Server Express is trimmed down, but it is an actual database engine (server application). Access is for people that don't know better.

understand this. friends biz started small and had very limited IT, so the data has grown over quite some time and now w/ greater resources, moving to the correct backend.

 

[Emulex]

sql server is just as easy to use as access - it just doesn't do UI.

you can pick up a sql server book (coming from access) and know SSIS in minutes.

now the GUI part - you need to program that yourself. .NET/PHP whatever. all on you.

 

[pitz]

AFAIK, all 15krpm drives are 2.5". They are just housed in 3.5" form factors for heat dissipation purposes.

 

[bob4432]

sql server is just as easy to use as access - it just doesn't do UI.

you can pick up a sql server book (coming from access) and know SSIS in minutes.

now the GUI part - you need to program that yourself. .NET/PHP whatever. all on you.

the software that is being used already supports sql, so it is just a matter of upgrading the db and a few other settings i believe.

 

[KentState]

How big is your database? If it's small enough, the entire thing may be cached in memory and the disk would just need to be fast enough to keep up with committed writes and log writes. Since the log is written sequentially, you really don't need anything more than RAID 10 on maybe 4 15k 72 GB drives. I would not mix a file server and SQL Server database files on the same disk. You will have very high disk queuing unless you have a lot of disk in a RAID 10 set.

 

[bob4432]

How big is your database? If it's small enough, the entire thing may be cached in memory and the disk would just need to be fast enough to keep up with committed writes and log writes. Since the log is written sequentially, you really don't need anything more than RAID 10 on maybe 4 15k 72 GB drives. I would not mix a file server and SQL Server database files on the same disk. You will have very high disk queuing unless you have a lot of disk in a RAID 10 set.

db is 1GB currently. also, this machine sees probably 15users max at one time, but not all are using the db, so i need to find out how many are using the software that uses the db at any 1 time too.

the more i read up on stuff, i don't think it will be an issue to run multiple arrays, i just need to verify the raid card will support it.

 

[KentState]

If the db is 1GB, then just dedicate however much RAM SQL Server wants after running it for a couple days. SQL Server only reads/writes directly to RAM during a query operation. SQLOS will then either read or write to/from disk either when a checkpoint occurs or the data is not in memory. I have a feeling that 4GB will be what you need to dedicate.