Tight rope walk

Usually, it is traumatic for a CIO to decide over a SAN implementation.

Having done that, and having obtained the funding for one, it is even more

traumatic to choose between an IP SAN or FC SAN. Both are luring. While the cost

of one is lucrative, the appeal of the other's high speed is enough to entice

an unsuspecting buyer.

Choice A Plenty



The decision to go for one or both or a combination of the two must be based

on the same factors that led the organization to decide to go for a SAN.

Vivekanand Venugopal, director, Software Solutions, APAC, Hitachi Data

Systems, says, “The intelligent buyer would use the application to dictate the

technology selection.” Customers can match the right applications to the right

technology, if they have clearly differentiated their key applications based on

performance, availability, cost, and manageability.

If the decision to adopt either FC or IP based SAN technologies is taken

logically, the organization will find itself going for a combination of the two.

The reason is simple. Rahul Gupta, vice-chairman, SNIA (he is also managing

director of Xserves) says, “If you talk of 20 TB storage on fiber channel, it

will cost you the moon.” So in architecture, the critical data can be put on

fiber channel, with FC drives; while the rest of the data can be in IP SAN with

SATA drives.

Fiber channel architecture is preferred when there is requirement in a local

environment for a very high-speed data transfer between the main network and the

storage network. But over a WAN or in a multi-location environment, where data

needs to be replicated, then probably an IP SAN could be a better option. This

is because TCP/IP is simpler protocol that is understood by almost all the

system administrators. Implementing fiber channel is a little more complex

besides being more expensive.

Anand adds, “When we do applications profiling for customers, we try and

understand what applications would be categorized. Say, as class A, which is

high performance, high availability, high recoverability, and very high response

times. Also, for these applications we will recommend a fibre channel-based

network. Because it is the most deterministic network available because of its

maturity.”

The Work Profile



Besides cost, speed is the primary difference between the two, and all other

differences flow from that.

The fiber channel standard achieves its high performance by assigning much of

the protocol processing to the dedicated hardware, a fiber channel host bus

adapter or a fiber channel switch. For applications of real-time data transfers,

such as e-commerce or online transactions, fiber channel is preferred.

While laying a fiber over a long distance may not be very feasible, the fiber

channel protocols can still be employed over WANs, by bridging the storage

solutions with dedicated bandwidth over leased lines or satellite connectivity.

It is important that the data carrying capacity of the media (fiber or

otherwise) be matched with the capacity of the HBAs and the FC switches.

Otherwise, the investment in them will be wasted.

IP SAN solutions are more useful where customers are looking for low-cost

solutions to block I/O applications, rather than file sharing applications. Some

of these block I/O applications could be deployment of exchange, SQL servers,

Linux-based database and backups. These are typically mission important, not

necessarily mission critical applications. They can function satisfactorily with

lower I/O and availability requirements.

Data and Traffic Analysis



Over the choice of use of FC SAN or an IP SAN, besides the needs of the

applications and the volume of data generated, the traffic analysis is also an

important factor. Thus, even if a branch office is generating huge amounts of

data, there may not be enough justification in putting up a fiber channel

solution there if the users of that data are sitting in the head office.

Whereas, if all this data is consolidated at one place and then viewed by a

large number of users, the FC SAN seems to appear viable.

What To Do For DR?



Today majority of the DR solutions have used IP circuit as a replication

mechanism. The cost of an FC circuit is as big a factor as the FC hardware, in

deciding to go for an FC solution. However, there is always a lag between the

primary and the secondary site. One solution to that is deploying a pre-datacenter

solution. According to Anand, many customers today say, “I would like to have

near zero loss and implement long distance replication as well.” The pre-data

center solution has a primary site and a secondary, which would be synchronous

over fiber channel over short distances. From the secondary site to the tertiary

site (which can be across states) will be over an IP circuit and an asynchronous

replication. With this kind of replication, near zero data loss implementations

can be achieved.

Some of the organizations that have implemented this type of pre-data center

solutions are HDFC Bank and Infosys.

Gupta says, “As far as loss is concerned, all vendors provide software to

ensure that the data packets that go from one place to another have been

acknowledged. So there is no question that data could get lost in transit.” In

DR kind of situations, if there is a file sharing kind of situation, and the

data has to be replicated in real time, an FC SAN over a dedicated leased line

bandwidth would be preferred. But if the replication can be done over a period

of time, then IP SAN will be fine. Another important issue is that from the same

database, or from the same storage network, how many people would simultaneously

access one type of data. If the number increases, the argument will shift

towards fiber channel.

Can a Fat IP Pipe compensate for an FC Solution?



No. Although it may imply that, it is not exactly like that. The difference

is that even if you have a huge pipe, the hardware on both sides of that pipe is

equally important. When talking about a video kind of application, a few frames

getting skipped won't matter. But if it is a critical online transaction, you

cannot afford jitter or loss of data. Since fiber channel is the more

deterministic of the two, for the second type of applications, it scores over

pure IP deployments. Though solutions like FCIP and IFCP run over IP media,

their protocols are different therefore their manageability would also be

different.

One major advantage of an IP SAN is that it can leverage the organization's

existing investment in the IP circuits and hardware, whereas, fiber channel will

require new and heavy investments. For this reason, SMEs are also driving the

demand for IP SANs, which have been picking up in the last one year.

Virtualization too does not seem to be taking sides between the two, allowing

the two to coexist. In an storage architecture, one can also have half portion

running on FC SAN and another half on IP SAN.

While in an environment an FC SAN can work over copper, it restricts the

capabilities of the fiber channel hardware. The SAN is a solution. Its fastest

speed is that of the weakest or the slowest link. So, the cost savings by

putting in copper have to be compared with the investment in expensive HBAs, FC

switches, and high- speed storage media and servers. Also, it requires

additional investments in the form of fiber channel extenders to be made. The

investment in fiber channel extenders is a big factor to keep while choosing to

route FC over IP.

On the whole, FC adoption is not determined only by the cost of circuits, it

is also determined by response times, round-trip response ties etc, which IP

cannot match, it is not that deterministic. That is where fiber channel

outscores.

At the same time, using fiber channel over IP is a more reliable

architecture, says Venugopal, because of its higher performance, compared to a

pure IP SAN.

By translating fiber channel control codes and data into IP packets, fiber

channel functions over IP. It is sometimes also called fiber channel tunneling

or storage tunneling.

A lot of the issues in the FC versus IP debate could be resolved in the

coming days with two simultaneous happenings. One, the cost of fiber channel

circuits is reducing, at the same time storage virtualization will allow system

administrators to manage the two seamlessly from one console. The two will form

part of one solution, without causing any pain to the administrators.

The Damages



Obviously in terms of absolute cost, IP SAN is less expensive than FC SAN.

Whether one takes HBA, or any other element that goes into the SAN, in case of

fiber channel it is extremely expensive as compared to whatever he uses in the

IP SAN.

Is There a Better Technology?



The test and R&D houses have been churning out reams of data on how FC

SAN has actually 'proven' to be less 'efficient' compared to iSCSI. But

that knowledge does not seem to have gone down to the market, with FC SANs still

remaining popular. One reason could be that a SAN is primarily an

enterprise-class solution. For lesser organization, a NAS may serve the

purposes. Of course, with growing business needs, even SMEs

need consolidated storage that can be easily managed. But, the

deterministic nature of fiber channel, and the reliability it has shown over the

years are still a big factor why the enterprises are going for it.

There is also an allegation that the popularity of FC SANs is vendor driven.

It is alleged  that the technology

is expensive. More importantly, many of the vendors are currently not ready with

a competitive IP SAN solution; they are pushing FC SANs to the customers. The

customers, forced with an artificially created lack of choice, have no option

but to go for what is made available by most of the vendors.

However, most vendors vehemently deny the charge. Venugopal says, “We

don't lead with the technology platform. Typically, we recommend a strategic

application optimized storage.” What this means is that customers view their

applications first and then optimize the storage infrastructure to meet the

storage requirements.

There are advantages that SAN can offer, advantages that may not have been

part of the original scheme of things in deciding to go for a SAN. Harish Shetty

of HDFC Bank says, “When we moved the applications from a DAS environment to a

SAN environment, we had to revisit the way we used to do our backups.

The process change actually allowed us to free the production servers for the

end of day activities.” This could happen because the backups now happen using

the off host server. While the bank's primary purpose of going in for a SAN

was for the purpose of protecting data by replicating the same to a DR site.

This also allowed the bank to reduce the amount of time for which the system was

taken down for the purpose of end of day activities. The bank, as Venugopal

pointed out, has a pre-data center solution, whereby it capitalizes on the

abilities of the SAN technologies, fiber channel as well as IP.

Alok Singh



aloks@cybermedia.co.in