One part queuing theory and 9 parts black arts, the design of high-end
systems is all about moving bottlenecks. For years the key bottleneck
was disk I/O. Then came RAID and the spotlight shifted to memory bandwidth.
Now with the new Intel Xeon chips sporting a 400MHz system bus and Hyper-Threading
architecture showing up in high-end workstations, as well as servers,
the bottleneck has spun full-circle back to disk I/O.
And once again the solution for boosting
disk I/O is RAID. The stumbling block is relative cost. With the cost
of all that silicon fire power at virtually the cost of sand, the cost
of a SCSI RAID solution for a small server or high-end workstation can
appear a bit pricey. That's why a number of vendors have turned to developing
proprietary controller technology to implement lower cost ATA-100 drives
in high-availability, commercial storage arrays.
ATA drives outnumber SCSI and
Fibre drives thousands to one. That's because their simpler controller
electronics keeps the price of ATA drives well below that of SCSI competitors,
As a result, ATA drives can be found in tens of millions of desktop
computer systems and a growing number of consumer electronic devices
such as personal video recorders.
| |
| openBENCH LABS SCENARIO |
UNDER EXAMINATION
Nexsan ATAboy ATA-based RAID appliance
8 IBM Deskstar 75GXP ATA100 drives
http://www.nexsan.com/
HOW WE TESTED
HP Netserver LP 1000r with 512MB RAM
http://hp.com
QLogic QLA12160 SCSI HBA
http://www.qlogic.com/
SuSE Linux 8.0
http://www.suse.com/
oblDISK v1
oblLOAD v2
KEY FINDINGS
Sequential data throughput on reads using the
Nexsan ATAboy (8 drives), consistently outperformed the
throughput measured on an Adaptec DuraStor 6200RS SCSI-based
RAID appliance (4 drives).
Transaction processing benchmarks using oblLOAD
demonstrated significantly lower results on the ATA-based
RAID appliance.
|
|
|
ATAboy and ATAboy2 RAID
appliances from Nexsan Technologies are two examples of the move
to ATA technology. The ATAboy, which openBench Labs tested, is
Nexsan's entry-level array family. The 4U ATAboy exports a single
LUN to a host system via an Ultra160 SCSI connection.
When it comes to all of the requisite
features for a serious storage array product, the Nexsan ATAboy
stands up well. The unit provides for on-line capacity expansion,
all active components are hot-swappable, and naturally there is
dynamic spare pooling of drives.
One feature, however, lets the Nexsan
ATAboy stand out from the crowd: a truly OS independent GUI for
configuration and monitoring. As a true appliance, the ATAboy
GUI can be accessed by any web browser on any platform. More often
than not, competitive products require that a Windows-based system
to run the GUI. This is not such a terrible inconvenience for
an IT server where there are numerous options for systems management
workstations; however, in the case of a high-end Linux workstation,
having to find a free windows system to check on your disk drives
can at the very least be a little off-putting.
The ATAboy2 offers a number of advanced
features that make its feature set more in line with such SCSI-based
products as Adaptec's DuraStor 6220SS. The 3U ATAboy2 incorporates
multi-LUN capability, dual host channels per controller, and will
support active-active fail over on the controller. In addition,
the ATAboy2 can be configured with a Fibre Channel interface for
use in a SAN.
|
|
|
To test the Nexsan ATAboy, we
installed a QLogic Ultra160 SCSI HBA into an HP Netserver LP 1000r
server, which was running SuSE Linux 8.0. From a separate
workstation, we configured the ATAboy RAID appliance.
Connecting to the appliance is
a trivial task from any browser. In line with its role as an entry-level
RAID appliance, the configuration menu is quite simple and free
from the plethora of options that no one would ever choose to
tweak.
Fundamentally, there are really only
three issues that must be answered. The first is the number of
logical volumes that will be presented to the host system. These
must be configured using discrete subsets of physical drives.
The next decision involves the type of RAID set—we chose to implement
RAID5 for our tests. Finally there is the matter of stripe size,
which we set to 64KB.
Our most difficult decision arose when
choosing the number of physical drives to include in our underlying
logical volume.
|
|
 |
| Nexsan's management GUI is completely
operating system neutral. It opens with a representation
of the physical environment including the state of power
supplies, fans, and drives. Mouse
over the environment view to see a more comprehensive
view of the logical RAID array. |
|
|
Typically, we configure SCSI-based RAID tests
with 4 physical drives. From a precise technical view point, for a direct
performance comparison to the Adaptec DuraStor 6200SR appliance, we
should have created two logical volumes on the ATAboy using the 8 IBM
Deskstar GXP75 drives. These drives sport SCSI-like performance specs
of 7200 RPM and sustained data rates of 37 MB/sec.
Nonetheless we chose instead to build a single
logical volume using all 8 IBM drives. The logic for this follows directly
from the logic of building an ATA-based RAID appliance. The lower cost
of ATA drives makes it possible to include more spindles in a RAID configuration
at the same price. Since the number of drives dramatically influences
performance, we chose to follow a more CFO-like price-performance approach.
What's more, the major performance issue for any ATA-appliance is in
dealing with a large number of drives, which is the performance hole
for ATA drives. We therefore conducted our tests on an 8GB ReiserFS-formatted
partition carved out of our logical 500GB drive.
We began our tests with our oblDISK benchmark,
which reads data sequentially in increasingly larger block-size requests.
We expected to see a slight edge for the Nexsan simply based on the
number of drives underlying the array. What we encountered was dramatically
improved read performance and a very slight degradation when performing
writes.
 
| |
With one thread spiraling through the
file and not getting extraordinary cache boosts, performance hovered
right at the sustained data rate of 80 MB/sec. specified by Nexsan.
This is right in line with the best of Ultra160 SCSI performance.
For a high-end graphics workstation or a simple file server, this
level of performance is exemplar.
Our next test, oblLOAD, probed at how
well the ATAboy appliance would perform in a transaction processing-centric
environment. In this test, which attempts to access 8KB blocks
of data in a database-like pattern (randomly within hot spots)
the ATAboy fared rather poorly. For this sort of environment the
more robust ATAboy2 is really necessary.
In a database-driven application with
hundreds of independent simultaneous users, the I/O pattern is
made up of a complex mix of localized high activity areas, such
as index tables, and essentially random access over the remaining
areas of the disk. In such a scenario, robust asynchronous I/O
is essential so as not to be held hostage by localized caching
performance. This is currently the one really bright spot for
Windows 2000 in any benchmark comparison with Linux—with a very
strong emphasis on currently. One of the hot areas of Linux kernel
development is to dramatically improve asynchronous I/O, and fortunately
the offending legacy constructs have all been easily identified.
Until these changes are implemented,
Linux transaction processing will remain bound by the speed of
cache hits. This puts a double whammy on an external RAID appliance.
In previous openBench Labs tests using PCI-based RAID controllers
such as the HP NetRAID-2M, performance lagged behind Windows 2000.
Nonetheless it was at a level that could sustain upwards of 1,000
requests per second and fulfill the requests in under 100 ms.
The big difference between the two is that the HP NetRAID-2M sits
internally on a 64-bit PCI bus while the Nexsan ATAboy is at the
end of an Ultra160 SCSI bus.
|
|
|
|
|
