Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

I am working with a postgresql database that sits on really old Hardware.

1 Sun Sunfire V60x with a single 3.0ghx Xeon FSB 400 512kb cache with 6gb of RAM running RHEL 4 (32bit)

The table space and logs are stored in the same 250gb LUN on a Hitach 9520v SAN running with 8 10k RPM USCSI drives with 146gb each.

Our budget is tight and given the performance advance of modern HDD's and cheap RAM I wonder how many new 15k RPM drives it would take to offer the same performance.

The table space is around 80 GB and our workload is probably 90% read and 10% write as the RDMS holds our organizations registry and is [almost] only used for reporting.

Edit: We will be using a brand new server with 2x Xeon 5620s and 32gb of RAM. I wonder if I can avoid the expense of the SAN.

share|improve this question
add comment

5 Answers

I'm not sure how much help more RAM would be as you have a 32bit environment. More RAM would be a pretty inexpensive upgrade to see how much it would help out.

New 15k drives would give you some performance improvement, depending on where your bottleneck currently is. If you aren't storage bound then spending money on the storage isn't going to do anything for you. I think you need to dig into the metrics of your server before you decide where to spend your money.

(My experience is from using Microsoft SQL Server on Windows, but the basic metric information and the underlying principals are the same.)

share|improve this answer
    
It will be a brand new server with 64bit OS. What I need to know is whether I can avoid the purchase of a new SAN and cover the I/O needs with directly attached HDDs. –  Gabriel T. Apr 14 '11 at 17:47
1  
What are your IO requirements? On the new server I'd buy as much RAM as possible. You can then attach the server to the existing array or use local disk. Local disk will often be faster than array storage, but you loose the manageability that you get with a storage array. What sort of latency are you seeing on the existing array and server? –  mrdenny Apr 15 '11 at 1:12
    
+1 for measuring and finding your bottleneck before purchasing new hardware. –  Ladadadada Nov 21 '11 at 20:57
add comment

Truthfully, the investment into HDDs would only give a slight improvement at best for disk-access. I think you'd be waisting most of your money if you invested in new drives. As far as RAM.... when it comes to RDBMS, recommended is "MORE". No matter how much you have, more is better. It wouldn't hurt to do some profiling either & find out where your pinch-points are. You might simply need to add an index to one or two fields & would see HUGE improvements in performance.

share|improve this answer
    
Well, were also planning to get a new server with 2 Xeon 5630's and 32GB of RAM. Will also move to Postgres 9 64bit.I just wonder if the IOPS offered by the Directly Attached Hard Drives will sufficiently replace that of the SAN currently in use. –  Gabriel T. Apr 14 '11 at 17:46
    
what kind of SAN? fiber-channel? iSCSI? directly attached? or are there switches/routers involved? are jumbo frames turned on? etc... lots of variables... directly attached drives will perform slightly better, there is no argument... but truthfully, once the database is cached in memory, the disk is only used when writing transactions. –  TheCompWiz Apr 14 '11 at 18:03
    
2gbps fibre channel. No switch involved, connected to the DB server with a single HBA. block level access. SAN is a Hitachi 9520v and each controller has 1gb of cache. 2 disks are reserved the the SAN OS and another 6 in a RAID-5 Array. It's not so much the SAN vs DAS performance delta that I am considering, but how much have drives advanced in the last 7 years IOPS wise given that there has been an at least 3-fold performance increase in desktop drives. –  Gabriel T. Apr 14 '11 at 20:09
    
Platter-based drives really have only gotten bigger. Unfortunately, the same laws of physics still dictate how fast you can move a head across a disk & how fast you can spin the platters before the whole setup gets unstable. There really haven't been great leaps in the number of IOPS a disk can perform unless you turn towards specialized drives.... (i.e. raptor drives & SSDs) –  TheCompWiz Apr 14 '11 at 20:45
    
I am obviously talking about 10k or 15k RPM SAS enterprise disks. This is a production environment. –  Gabriel T. Apr 14 '11 at 21:19
add comment

I guess that you could at least match the performance of your old SAN with a modern server grade SSD (or a few with a good RAID controller), and likely have much better access times.

share|improve this answer
add comment

SAN is used to export block devices .. the only thing that SAN offers is flexibility in storage (single point of storage, any point of mount over the net of connection of you choosing). In speed you cannot beat the local attached storage in hardware (or software) raid ... if you need speed, the easiest, fastest and cheapest (as you already have the hdds) solution would be (IMHO) raid 10 (4 hdds in raid1 and the both volumes in raid0) and that would give you 2 x 146 gb of storage ... also 2 X hdds in raid 1 and 4 x raid1 in raid0 would also be a solution... anyway at this stage and at this size i would strongly recommend to look over ssd prices and capacities .. 2 X 240 gb ssd (consumer) in raid1 are smoking fast!

share|improve this answer
    
I've considered SSD's but I've read that the only way to avoid data corruption in case of a power outage or other failure is to disable the on board cache and only use the cache of the raid controller. Supposedly that degrades their performance dramatically and messes with their wear leveling algorithms. –  Gabriel T. Apr 14 '11 at 20:32
    
i really doubt that! every manufacturer have something in place to prevent that, and many more on enterprise level ssds (like some big capacitor). For hardware raid always is good to have backup battery ..but also if we are talking about this that means that you already have an GOOD ups in place, and the server will shutdown itself when battery is critical. the ups is the first line of defense against data corruption .. or maybe second after S.M.A.R.T. monitoring ;) –  adrian_sev Apr 14 '11 at 22:26
add comment

Have you thought about software based ISCSI SANs as an alternative, check http://www.starwindsoftware.com/starwind. You will be able to cut cost, have your centralized storage, and have a scalable solution for clustering if that is something you are considering to do in the future. Otherwise, direct attached storage will probably satisfy your I/O.

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.