Tweet
Bio-It World Expo 2009: The Dagdigian Doctrine
By Kevin Davies
April-May 2009
Bio-ITworld.com
http://www.bio-itworld.com/ 04/28/09/bio-ITWorld-Keynote-Dagdigian-doctrine.ht ml
April 28, 2009 - BOSTON - In the opening keynote of the 2009 Bio-IT
World Conference & Expo, Chris Dagdigian delivered a candid assessment
of the best, the worthwhile, and the most overhyped information
technologies (IT) for life sciences.
Some formerly hyped technologies were now mainstream, including
virtualization and storage. Others were over hyped but still on
balance worthwhi le, such as green IT and utility computing. And in
the future, Dagdigian saw major benefits accruing from trickle-down
best practices and federated storage.
Dagdigian, a founding partner for the BioTeam IT consultancy, is a
regular speaker at the annual Bio-IT World Expo. This year, he brought
his trademark `trends in the trenches' talk to the opening plenary
session. Offering his customary disclaimers, he said he was
`comfortable with deliverables, but not comfortable with being a
talking head or a pundit.' The audience begged to differ.
Already mainstream
Dagdigian divided his talk into three sections: reviewing what he
called old news,' discussing currently exciting' technologies, and
those that could be exciting in the future. But, he warned, there is a
ridiculous list of stuff on the `hypemeter.'
In the already mainstream' category, Dagdigian tagged virtualization
and the Cb io-IT storage tsunami.' Most of his recent work was in
helping `people getting buried by instruments.' The community had been
talking about the data deluge' for the past four years. It was still
buried,' said Dagdigian, but the problem domain is understood. Our
smallest customers aren't losing hope, and our biggest customers are
staying ahead.'
Last year, Dagdigian saw the first 100-Terabyte (TB) single namespace
project.' But, he cautioned, he had also witnessed for the first time
a 10 TB catastrophic data loss' with consequent job losses. The loss,
which occurred in a g overnment lab, resulted primarily from
double-disk failure in a RAID5 volume holding SAN F5
metadata. Dagdigian said he used to be huge fan of RAID5,' but
abandoned it last year. The statistical probability [of failure] is
too high - it's going to happen!' Everything is now RAID6, with
maximum attention paid to monitoring and maintaining data integrity.
In a welcome development, data triage discussions were spreading
beyond cost-sensitive industry organizations. Everyone has come to the
realization that data triage is a given,' said Dagdigian. Displaying
his favorite slide, Dagdigian showed a screen short of an 82-TB folder
on a Mac. Even more impressive was a 1-Petabyte (PB) output mounted on
a Linux system output.
While storage was cheap (and getting cheaper), operational costs,
staff, tape and backup costs were still fairly constant. Users,
enamored with the plummeting price of TB storage devices from C ostco,
did not understand enterprise IT and backup requirements. IT
organizations need to set expectations, because the electronics market
is skewing expectations,' said Dagdigian. Many next-generation
sequencing grants simply didn't budget for storage, let alone a 100+
TB storage system. Meanwhile, Dagdigian noted, the Broad Institute
already has more than 1 PB of storage.
Unlimited data storage is over,' noted Dagdigian. It's simply not
possible to back up all data, keep it safe, secure, and so
on. Sometimes,' he said, it's better20to go back to the -40 F freezer
and repeat the experiment.'
In short, storage was no longer a major bottleneck - rather, that fell
to chemistry, reagents, and human factors. Customers were starting to
trust instrument vendor software more. 'The problems are not as scary
as they once seemed,' he said. Dagdigian also noted that storage
devices are running more 3rd party software, such as Ocarina Reader
software on Isilon, and Ocarina Optimizer on BlueArc.
Virtualization offered the lowest hanging fruit,' said Dagdigian. The
tipping point, Dagdigian said, was the live migration of a VMS
(virtual memory system) without requiring a proprietary file system
underneath. In 2009, he helped build and design=3D2 0a virtual
collocation facility for an academic west coast campus, which was
experiencing limits imposed by electrical power and air
conditioning. 400 servers are currently virtualized on a lightweight
simple platform,' said Dagdigian. Large numbers of physical servers
had been shut down, realizing significant savings from de-duplication,
compression and thin provisioning, not to mention
electricity. Moreover, scientists now had full administration control.
Coming soon, said Dagdigian, virtualized cluster head nodes.' Not
coming soon: grids and clusters distributing entire VMS for tas k
execution. It wasn't practical, argued Dagdigian but rather a case of
marketing winning out over practical stuff.'
Green IT
In the category of hyped beyond all reasonable measure - but still
worth pursuing,' Dagdigian said green IT could deliver real electrical
savings. Use green IT for political cover,' he urged the audience. A
deployment of a Nexan SATABeast had led to a 30% reduction in power
draw with no impact on cluster throughput. One of his best 2009
moments came in deploying a Linux HPC cluster for a west coast
organization. The system interface talks to Platform LFS, powering
nodes up and down and sending automatic email alerts to management
such as: Hello, I've saved $80K in facility costs this year.'
Utility (or cloud) computing was not rocket science, but fast becoming
mainstream,' said Dagdigian. Amazon web services is the cloud,' he
said. It's simple, practical, and understandable,' and enjoyed a
multi-year head start on the competition. The rollout of features was
`amazing,' such as Hadoop and applications for short-read sequence
mapping. I drank the EC2 Kool-Aid: I saw it, I used it, I solved
real-world problems,' said Dagdigian.
The biggest problem was ingesting data into the cloud. There is no
easy solution,' he said. How does one push 1TB/day into Amazon? Have
patience,' said Dagdigian, expressing 100% confidence' that Amazon is
working on the problem. If we can solve data ingestion problem, I see
a lot of scientific data taking a one-way trip into the cloud. Data
would rarely, if ever, move back. If I take it back, it's going to be
really obnoxious -- big data pipes or people driving minivans of USB
drives.'
Worth Watching
In his worth watching' category, Dagdigian cited federated storage and
the trickle down of best practices from Amazon, Google and others.
Amazon, Google and Microsoft had all been computing at such a scale
that it was too much of a trade secret, he complained. But there were
signs, such as a recently released video of the Google data center
circa 2004, that their best practices will eventually trickle out,
benefiting the entire community.
By Kevin Davies
April-May 2009
Bio-ITworld.com
http://www.bio-itworld.com/ 04/28/09/bio-ITWorld-Keynote-Dagdigian-doctrine.ht ml
April 28, 2009 - BOSTON - In the opening keynote of the 2009 Bio-IT
World Conference & Expo, Chris Dagdigian delivered a candid assessment
of the best, the worthwhile, and the most overhyped information
technologies (IT) for life sciences.
Some formerly hyped technologies were now mainstream, including
virtualization and storage. Others were over hyped but still on
balance worthwhi le, such as green IT and utility computing. And in
the future, Dagdigian saw major benefits accruing from trickle-down
best practices and federated storage.
Dagdigian, a founding partner for the BioTeam IT consultancy, is a
regular speaker at the annual Bio-IT World Expo. This year, he brought
his trademark `trends in the trenches' talk to the opening plenary
session. Offering his customary disclaimers, he said he was
`comfortable with deliverables, but not comfortable with being a
talking head or a pundit.' The audience begged to differ.
Already mainstream
Dagdigian divided his talk into three sections: reviewing what he
called old news,' discussing currently exciting' technologies, and
those that could be exciting in the future. But, he warned, there is a
ridiculous list of stuff on the `hypemeter.'
In the already mainstream' category, Dagdigian tagged virtualization
and the Cb io-IT storage tsunami.' Most of his recent work was in
helping `people getting buried by instruments.' The community had been
talking about the data deluge' for the past four years. It was still
buried,' said Dagdigian, but the problem domain is understood. Our
smallest customers aren't losing hope, and our biggest customers are
staying ahead.'
Last year, Dagdigian saw the first 100-Terabyte (TB) single namespace
project.' But, he cautioned, he had also witnessed for the first time
a 10 TB catastrophic data loss' with consequent job losses. The loss,
which occurred in a g overnment lab, resulted primarily from
double-disk failure in a RAID5 volume holding SAN F5
metadata. Dagdigian said he used to be huge fan of RAID5,' but
abandoned it last year. The statistical probability [of failure] is
too high - it's going to happen!' Everything is now RAID6, with
maximum attention paid to monitoring and maintaining data integrity.
In a welcome development, data triage discussions were spreading
beyond cost-sensitive industry organizations. Everyone has come to the
realization that data triage is a given,' said Dagdigian. Displaying
his favorite slide, Dagdigian showed a screen short of an 82-TB folder
on a Mac. Even more impressive was a 1-Petabyte (PB) output mounted on
a Linux system output.
While storage was cheap (and getting cheaper), operational costs,
staff, tape and backup costs were still fairly constant. Users,
enamored with the plummeting price of TB storage devices from C ostco,
did not understand enterprise IT and backup requirements. IT
organizations need to set expectations, because the electronics market
is skewing expectations,' said Dagdigian. Many next-generation
sequencing grants simply didn't budget for storage, let alone a 100+
TB storage system. Meanwhile, Dagdigian noted, the Broad Institute
already has more than 1 PB of storage.
Unlimited data storage is over,' noted Dagdigian. It's simply not
possible to back up all data, keep it safe, secure, and so
on. Sometimes,' he said, it's better20to go back to the -40 F freezer
and repeat the experiment.'
In short, storage was no longer a major bottleneck - rather, that fell
to chemistry, reagents, and human factors. Customers were starting to
trust instrument vendor software more. 'The problems are not as scary
as they once seemed,' he said. Dagdigian also noted that storage
devices are running more 3rd party software, such as Ocarina Reader
software on Isilon, and Ocarina Optimizer on BlueArc.
Virtualization offered the lowest hanging fruit,' said Dagdigian. The
tipping point, Dagdigian said, was the live migration of a VMS
(virtual memory system) without requiring a proprietary file system
underneath. In 2009, he helped build and design=3D2 0a virtual
collocation facility for an academic west coast campus, which was
experiencing limits imposed by electrical power and air
conditioning. 400 servers are currently virtualized on a lightweight
simple platform,' said Dagdigian. Large numbers of physical servers
had been shut down, realizing significant savings from de-duplication,
compression and thin provisioning, not to mention
electricity. Moreover, scientists now had full administration control.
Coming soon, said Dagdigian, virtualized cluster head nodes.' Not
coming soon: grids and clusters distributing entire VMS for tas k
execution. It wasn't practical, argued Dagdigian but rather a case of
marketing winning out over practical stuff.'
Green IT
In the category of hyped beyond all reasonable measure - but still
worth pursuing,' Dagdigian said green IT could deliver real electrical
savings. Use green IT for political cover,' he urged the audience. A
deployment of a Nexan SATABeast had led to a 30% reduction in power
draw with no impact on cluster throughput. One of his best 2009
moments came in deploying a Linux HPC cluster for a west coast
organization. The system interface talks to Platform LFS, powering
nodes up and down and sending automatic email alerts to management
such as: Hello, I've saved $80K in facility costs this year.'
Utility (or cloud) computing was not rocket science, but fast becoming
mainstream,' said Dagdigian. Amazon web services is the cloud,' he
said. It's simple, practical, and understandable,' and enjoyed a
multi-year head start on the competition. The rollout of features was
`amazing,' such as Hadoop and applications for short-read sequence
mapping. I drank the EC2 Kool-Aid: I saw it, I used it, I solved
real-world problems,' said Dagdigian.
The biggest problem was ingesting data into the cloud. There is no
easy solution,' he said. How does one push 1TB/day into Amazon? Have
patience,' said Dagdigian, expressing 100% confidence' that Amazon is
working on the problem. If we can solve data ingestion problem, I see
a lot of scientific data taking a one-way trip into the cloud. Data
would rarely, if ever, move back. If I take it back, it's going to be
really obnoxious -- big data pipes or people driving minivans of USB
drives.'
Worth Watching
In his worth watching' category, Dagdigian cited federated storage and
the trickle down of best practices from Amazon, Google and others.
Amazon, Google and Microsoft had all been computing at such a scale
that it was too much of a trade secret, he complained. But there were
signs, such as a recently released video of the Google data center
circa 2004, that their best practices will eventually trickle out,
benefiting the entire community.