[THIN] Re: OT: Effect of Solid State drives
- From: "Malcolm Bruton" <malcolm.bruton@xxxxxxxxxxxxxxxxxx>
- To: <thin@xxxxxxxxxxxxx>
- Date: Wed, 6 Jun 2007 19:05:06 -0400
Rick. Do you have any details of the performance? Did a quick Google
but didn't see much. I've read that at the moment there is no real
difference between speed of the flash drives vs. traditional laptop hard
disks. I do expect this to change though in the future. The new 2.5
inch 10,000 RPM SAS drives are pretty fast vs. the old 3.5 inch 15 K
SCSI drives and I know that HP/Seagate are talking 15K 2.5 inch SAS
versions soon as well.
The article was in a PC mag in the UK but I can't find an online
version.
The Samsung drive was faster in some benchmarks when compared with a
5400 std notebook drives but not all tests
Malcolm
From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On
Behalf Of Rick Mack
Sent: 06 June 2007 22:54
To: thin@xxxxxxxxxxxxx
Subject: [THIN] OT: Effect of Solid State drives
Hi,
Pardon this off topic post, but I think there are ging to be some
interesting changes in our server environment with the advent of almost
affordable solid state flash dirves (SSD)
Both Samsung (in production) and SANDisk (near production) have
announced 64 GB solid state drives in a 1.8 and 2.5 inch form factor.
I thought it'd be fun to comment and speculate on some of the things
that are likely to happen.
The main push for these drives is into the laptop market, but I think
the impact will be much greater once people realise that they will solve
major server bottlenecks.
At the moment they'll probably cost about $10-15 a GB but considering
the total cost of your average server, that's really not going to break
the bank.
With 100 microsecond "seek" times and sustainable data transfer rates of
45 MB/sec (now) they are going to blow small (72 GB) server disks out of
the water. Very fast seek times means no seek optimization is required
so the performance difference between parallel IDE, serial IDE and SCSI
(serial or parallel) for random i/o will be irrelevant.
Imagine an IDE-based blade that blows a SCSI blade out of the water.
Or a dual quad core server that can actually use all that CPU
horsepower.
Smaller and more reliable power supplies because we don't have to deal
with disk startup surges.
Hard disks that can be treated like USB flash drives.
A 1000000 hour MTBF, "instant" spinup time means very fast and reliable
disk storage systems can can start being power optimized.
SANs will be slow compared to local hard disks and no more reliable.
Considering the cost per GB of your average SAN and the observation that
if we disregard the management software their main claim to fame is
reliability, I think it's fair to speculate there will be a shakeup in
the SAN market. The main claim to fame for SANs is their reliability,
but what if we've got an even more reliable technology that's much
faster. An SSD based SAN would be sufficiently simple that it shouln't
cost a lot more. Smaller, faster, quieter.
Bank 10 SSD (1.8") drives together with RAID 0 and we're talking about a
disk subsystem that can maintain disk i/o at nearly data bus speeds.
Lets put 50-100 SSD disks into a SAN and suddenly a lot of things have
to change.
Fibre channel won't be fast enough and using a traditional switch fabric
just won't hack the pace. We're likely to see SAS or SATA direct connect
options into a SAN, with as many SAN connect interfaces as we have
SAN-requiring hosts. Or a distributed SAN infrastructure, maybe based on
iSCSI.
Traditional hard disks will only be used for temporary or very high
capacity storage. Unless Seagate and Maxtor produce some interesting 3
layer (RAM cache, Flash Cache, disk) hybrid technology to make up the
performance difference, their shares are going to plummet.
I can hardly wait ;-)
regards,
Rick
--
Ulrich Mack
Commander Australia
Other related posts:
