Re: [ADSM-L] Ang: Re: DB2: SSD vs more RAM
2010-11-23 02:29:29
With limited budget and spinning disks you often have to choose security only.
But it's somewhat an illusion, since disks are so slow that your system really
doesn't work: your expiration cycles take days to complete (at least with
fragmented db - maybe TSM6 is much better), and doing many simultaneusly
restores (tons of small files) during disaster recovery takes forever when your
db disk setup is the bottleneck.
Short stroking Intel MLC SSD:s makes them last longer specifically with random
writes. Of course they still wear out faster than SLC, so heavy database
workload propably would be too much. (But they are used for many server
workloads) With SLC it's almost a non-issue.
Data loss is never acceptable, but for some organizations outages just might
be. After all, many run TSM in single server without any H/A clustering.
Anyway I have to agree that if your db is 200GB and budget allows six
performance disks, go with disks. if the DB would be 50 GB and same budget
(let's say $1500) I would not even think using disks instead of SSD.
------- Original message -------
From: Daniel Sparrman <daniel.sparrman AT EXIST DOT SE>
To: ADSM-L AT VM.MARIST DOT EDU
Sent: 2010-11-23, 1:01
Skipping raid/mirroring is probably the worst thing he could do. When the
database hits that size, you can expect your organisation to want the TSM
server up at all times. A disk outage would not really meet that requirement.
SSD disks only have a longer lifespan during lots of reads / less writes. In a
TSM environment, that wouldnt be true, thus not giving the SSD's a longer
lifespan.
a) Secure your TSM server, it's your lifeline whenever everything else goes
wrong
b) Go for performance
Never turn those 2 points the other way around.
Regards
Daniel Sparrman
-----"ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU> skrev: -----
Till: ADSM-L AT VM.MARIST DOT EDU
Från: Henrik Ahlgren <pablo AT SEESTIETO DOT COM>
Sänt av: "ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU>
Datum: 11/22/2010 23:27
Ärende: Re: DB2: SSD vs more RAM
Yep, doing a 200 GB database with high-end and reduntant SLC NAND is definedly
not cheap (let alone the 2 TB Bill Colwell described in his post). Not that
bunch of 15K disks and the power to run them is free, either. Cost per IOPS for
disk is actually terrible.
Just a thought (don't take too seriously!): what if you'd be willing to take the risk and
forget RAID/mirroring, after all solid state is pretty reliable these days. Of course -
in addition to perfect DB backup strategy you need anyhow - put your transaction logs to
different disk: spinning disk is great for that (it's more or less sequential I/O). And
maybe even use cheaper MLC NAND - if you "short stroke" (google for intel ssd
overprovisioning) a 160 GB X25-M down to 128 GB, you get three times the endurance, so it
should last for quite a long time even with DB workload. Of course, like tapes, you have
to treat NAND media as consumables, and keep an eye on the S.M.A.R.T. media wearout
indicators.
Yeah, too radical and risky for most. We'll just have to wait for couple of
years more to finally get rid of rotating rust for random IO usage once and for
all.
On Nov 22, 2010, at 7:13 PM, Pretorius, Louw <louw AT sun.ac DOT za> wrote:
Well as I was specing a new TSM server i thought, why not try for the best
performance possible and although the SSD drives drives the server costs up by
50% it wasn't out of the ballpark, therefore I wanted to hear from the
community what their ideas were.
As it stands I have a 100GB DB currently ~50% used but according to IBM TSM 6.2
will require double DB size hence 200GB and since we are expecting a 40%
data-growth next year and will be implimenting Dedupe I thought why not see how
the price/performance goes on other sites.
As db2 is a fully featured DB I thought that the alternative would be to give
it more RAM, as it's so much cheaper than SSD's. And also how I could
configure my DB2 to use the extra RAM that I will be throwing at it in any case.
With the current feedback I will be sticking to 6 x SAS 15K and 24GB RAM...
Please if there's any other opinions let's hear it, the more opinions the more
wisdom...
Regards
Louw Pretorius
-----Original Message-----
From: ADSM: Dist Stor Manager [mailto:ADSM-L AT VM.MARIST DOT EDU] On Behalf Of
Henrik Ahlgren
Sent: 22 November 2010 11:25
To: ADSM-L AT VM.MARIST DOT EDU
Subject: Re: [ADSM-L] DB2: SSD vs more RAM
Or maybe he has a huge amount of DB entries? If his options are either six SAS
15K or eight SSDs (50GB each), it means his DB is propably in the multi-hundred
gigabyte range. If he just needs the IOPS for smaller DB, then he would not
need 8 SSDs to beat 6 platters, even one or two could be enough. (Just one
Intel X25E does 35K IOPS random 4K read.) I'm not sure how much doubling the
RAM would help with operations such as expiration, DB backup etc. compared to
nice SSD setup.
I'm wondering why so little discussion here on using solid state devices for
TSM databases? Some of you must be doing it, right?
On Nov 17, 2010, at 7:50 PM, Remco Post wrote:
SSD to me seems overkill if you already have 24 GB of RAM, unless you need
superfast performance and are going to run a very busy TSM server with a huge
amount of concurrent sessions.
--
Gr., Remco
On 17 nov. 2010, at 12:16, "Pretorius, Louw <louw AT sun.ac DOT za>" <louw AT
SUN.AC DOT ZA> wrote:
Hi all,
I am currently in the process of setting up specifications for our new TSM6.2
server.
I started by adding 8 x SSD 50GB disks to hold OS and DB, but because of the
high costs was wondering if it's possible to rather buy more RAM and increase
the DB2 cache to speed up the database.
Currently I have RAM set at 24GB but its way cheaper doubling the RAM
than to buy 8 x SSD's Currently I have 8 x SSD vs 6 x SAS 15K
--
Henrik Ahlgren
Seestieto
+358-50-3866200
--
Henrik Ahlgren
Seestieto
+358-50-3866200
|
|
|