How did he do it? Dataquest chief editor Prasanto K Roy finds out in a
tête-à -tête with this Renaissance man in Mumbai. Once a genuine cowboy, Hitz,
currently executive VP, engineering of Network Appliance maintains that he got
"valuable management experience herding, branding and castrating cattle
before taking up his present career"! He spoke at length on a wide range of
topics, from the future of Storage Area Networks (SAN), storage virtualization,
to bandwidth requirement for storage consolidation. Excerpts:
On how the company came to be called Network Appliance and the ''storage
appliance'' concept.
In 1985, Sun was number one provider of TCP/IP routing systems with its
workstation solutions. Then Cisco decided to build special-purpose dedicated
devices that would only do routing as they thought TCP/IP was here to stay.
Hence, the concept of appliance was introduced successfully for the first time.
When we named our company Network Appliances, we copied Cisco''s reasoning, but
from the perspective of storage. We thought storage over networking was here to
stay and our goal was to successfully emulate Cisco''s success with the
appliance concept.
|
At that time, no one understood the term appliance. In fact, all VCs
therefore refused to invest in NetApp. However, we were determined to follow
Cisco''s model, and incidentally, the chairman of the board of directors at
Cisco became the chairman of our board of directors. After that, we also
obtained VC funding from Sequoia. We thought we would come out with a whole
bunch of appliances, and so named our company Network Appliance instead of
Storage Appliances. Generally, if you take a bird''s eye view, the concept of
appliances has succeeded for storage-as it has with us, or networking routers
have in the instance of Cisco, or networking printers as in HP''s case. It has
failed whenever something like a general-purpose server appliance has been
tried, because that makes for unnecessary complexity.
It seems to be that network speeds and storage channels
are converging and might even intersect at some point of time. Does this imply
that there might not be any compelling reasons to move to SAN?
There are at least three compelling reasons why SAN fiber channels will stay
viable at least for the time being. A look at the performance numbers for both
Ethernet and fiber channel, will tell you that they lie very close within the
same bandwidth. But, it is on the reliability front that fiber channel scores.
What CIOs like about fiber channels is the QoS model and the way the packets are
handled. Ethernet is improving, but it still has a long way to go. The next
reason is that most enterprises would rather have a separate dedicated network.
In many companies, even application- and storage-group people are not allowed to
build TCP/IP networks. The diktat is that if it is a TCP/IP network it has to be
provided by the network infrastructure group. So if the storage group has to
build a dedicated network, it has to be on fiber channel. Lastly, adoption of
fiber channels has a lot to do with the certifications in the industry. Many of
the high-end applications like databases or other enterprise apps are certified
only on fiber channel.
My prediction is that SANs might become redundant around
2006-07 when 10 gigabit Ethernet will go mainstream. Enterprises will then need
to move their SAN fiber channel to10 gigabit fiber channels and the additional
cost involved will make many of them to go for a single Ethernet deployment. We
feel that there will be a large deployment of even iSCSI and NAS, which is
happening anyway in the current scenario.
Would you therefore expect a movement back from SAN to
Ethernet-based storage solutions?
I would not look at the issue in that way. What happened was when most SAN
implementations took place about two years ago, the network infrastructure in
most cases was not that good. While Ethernet has improved considerably, we might
not see a reverse exodus from SAN. In two to three years, when gigabit Ethernet
becomes the norm, we might see Ethernet becoming the preferred mode over fiber
channel. We might even have a solution like iSCSI that straddles the worlds of
both Ethernet and fiber channel.
How do you balance the choice of technology between a
SAN/NAS and the access mechanism?
One of the reasons why wires are becoming less and less important is because
the wire speed of Ethernet has begun to approximate more and more the wire speed
of fiber channels. The other reason is that, while fiber channel was much faster
than 100base Ethernet, there were no performance dips with respect to most
applications, because the bottleneck was the disk drive, not the wire. The
answer to the problem of reducing access mechanism bottlenecks is storage
consolidation, whereby grouping multiple drives will lead to enhancement of
performance. But access mechanisms or disk drives should not be the determining
factor in the choice of NAS or SAN.
Most storage vendors claim that even if storage across the
enterprise lies in disparate devices, by virtue of storage virtualization it can
be used/managed as a single logical unit. What are the problems here?
Whenever someone talks about storage virtualization, you need to ask: what
is not good about the available storage solution that you have to talk about
virtual storage in the first place? The first problem is that it is not big
enough, so you cannot solve all your problems with one storage solution, but
instead, require multiple solutions. To make them look like one solution, you
can put them in one network like NAS/SAN and this would also require management
tools. But the question remains: how can we take little chunks of storage and
make them look bigger? Ironically, the second problem is that virtual storage is
too big. For instance, for 100 different users, you need to buy just one, big
storage system instead of multiple devices, because it works out to be cheaper.
And then you will make this one look like tiny systems through the use of
different management tools. The third challenge is how to handle these
management tools, since this would involve not only applications from storage
vendors, but even apps like HP Openview or CA Unicenter that are basically
infrastructure management tools. Handling so many storage-management as well as
infrastructure tools can become cumbersome.
With bandwidth being expensive and scarce in India till
quite recently, for many enterprises the WAN architecture across cities cannot
support technologies like storage consolidation...
The Indian situation is quite similar to the one in the US, where bandwidth
became cheaper and more easily available after the telecom operators
deregulated. But even then there are still very few enterprise applications that
run remotely. So even for storage consolidation, you will not find a situation
where enterprises consolidate all of their storage in one single city. Even in
the US there is not enough bandwidth, and I do not expect the situation to be
any different in India. But at least each of the home directories on individual
desktops, which have apps like PowerPoint and Word, can be consolidated to a
single site. Plus, enough bandwidth is now available to have an online
consolidated mirror in another city, which can act as the DR site.
Rajneesh De with
Prasanto K Roy