Wintel-Based Rackmount Servers - The Big Get Bigger
R.A. Krause - August, 1999
high-end rackmount ("RM") Intel-based server market is defined as servers (usually
more than one) mounted into floor-standing ("FS") equipment racks, plus storage
and other peripherals. Prices range from ~$10K for a bare-bones configuration
to $100+K for a "loaded" configuration. This market grew out of the need for
customers to put more equipment in limited floor space, so that systems grow
up first, then out. (Note that this analysis covers only multi-system configurations;
it does not cover mainframe-style systems, in which a single system fills one
or more racks.)
market size for all rackmount servers is estimated at ~$5 billion, and is estimated
to grow 40%-60% (to $7-8 billion) in the next 1-2 years [Source: TEC]. The expected
rapid growth within the segment is due to two main factors: the historical trend
of increased server shipments as a whole, and a shift from floor-standing (also
called "pedestal", "tower", and "mini-tower") servers to rackmount units. With
the exception of "internet appliance"-like servers (driven by ISPs and similar
internet-related businesses), the shift will be more pronounced in the high-end
markets than in the low-end and mid-range market. Thus, it can be considered
a predatory market, its prey being the floor-standing market.
with the server market in general, this market appears to be consolidating.
Presently there are only four vendors with significant (10% or more) market
share. Consolidation is expected to continue.
markets are clusters - traditionally built using numerous free-standing systems.
"Cluster in a box" appears to be a market which will grow significantly in the
coming years, in some cases supplanting the current many-box configurations.
current market share leader is Compaq (~30%), followed by Dell, HP, and IBM
(~10-15% each). Compaq's leadership stems from its historical dominance in the
server marketplace, coupled with quality products. In addition, Compaq offers
the greatest diversity of rackmount products, and has overcome the problem of
a 42U rack that won't fit (upright) through a door. This provides a significant
advantage, because it helps allow Compaq (or its resellers) to ship racks with
systems already installed, without worrying if the customer needs "ten men and
a boy" to tip it over just to get it through a door.
is having problems trying to shift from a VAR/channel-based distribution model
to a direct sales model (similar to Dell's). In addition, there have been complaints
regarding the difficulty of building and installing racks and servers - customers
don't want to spend an excessive amount of time installing systems, nor do they
want to be forced into hiring an expert to do it for them. Compaq has recently
redesigned their racks, but it is too early to tell if the complaints have been
addressed. A potential weakness is, paradoxically, its breadth of offerings
- if there is no clear distinction between products, confusion can arise. For
example, Compaq presently offers five distinct 4P models.
virtue of their current position in the PC server market, the current market
challengers are Dell, HP, and IBM. All offer rack solutions (servers and racks)
similar to Compaq's offerings.
Dell's strengths are its price, delivery, customer satisfaction, and its
aggressive pursuit of Compaq. Its weaknesses are in server management (incl.
remote mgmt.), relying on HP's OpenView in this area.
HP's strengths are its system design, and its reputation for quality. Its
weakness is its "incomplete" product set: Compaq and Dell have 4U/4 CPU
systems, whereas currently HP's smallest 4 CPU system requires 7U in the
IBM continues to try to make inroads in the high end, and the Netfinity
7000 M10 has won numerous magazine and show awards, as well as performing
extremely well in various benchmark tests. However, IBM presently lacks
a 4P/4U server to compete with Compaq and Dell.
to the consolidation taking place in the Intel-based server industry - the big
four are controlling an increasingly greater percentage of sales volume (~75%
in June '99 vs. 66% in 1998 vs. 58% in 1997) [Source:IDC]. In this marketplace,
the "also-rans" will be those vendors who are not in the top four in sales (e.g.
Data General, Unisys, Toshiba, Micron/NetFrame), or those whose names are not
associated with either datacenters/data warehousing, clusters or ----similar
market will continue to grow, both in size and market share, over the next 3-5
years. This market share increase will come at the expense of freestanding systems.
Market share split will change from 40%-50% RM vs. 50%-60% FS (now) to 70+%
RM vs. 30% FS within three years. In raw numbers, this is expected to translate
into $10 billion per year in 2001, showing a CAGR of approx. 30%.
"cannibalization" of the freestanding server market will be due to the desire
to increase processing density per square foot of floor space. In addition,
customers want the ability to add servers without having to rearrange their
computer room. Although the initial cost for installing a rack system is a little
high for some users (>$1000 for a typical rack), customers are willing to
make the tradeoff for increased flexibility for the future.
the top four vendors: Compaq will retain its lead and even increase its overall
share, but will lose 3-5% of its lead over Dell in the next two years. Dell
will continue to hold the #2 position, but will gain market share at the expense
of Compaq and the small players. HP and IBM will maintain their relative positions.
need to continue to provide more functionality per unit volume - "density
need to provide product breadth - a solution for all product segments in
the rack server spectrum.
can differentiate themselves by making installation and configuration simple
and easy (across a number of areas). One way this can happen is for vendors
to pre-build the systems, so that the customer doesn't have to go through
the pain and expense of getting the racks built on-site.
long-term winners will be the already-large vendors. Market consolidation has
already started, and is expected to continue. The long term losers will be the
smaller vendors, e.g. Unisys, Data General, Toshiba. Although it is unlikely
(<10% probability) that the majority of smaller players will disappear completely
in the next five years, market share for the "big four" is expected to increase
(80% probability). To move into the large-market-share crowd, smaller vendors
will need to provide something extra, whether a significant new technology,
or attacking a previously-overlooked market. However, given the relative maturity
of the market and technology, and that Intel's technology drives all the vendors,
we do not believe there will be a technological breakthrough significant enough
to accomplish this. (80% probability of non-breakthrough)
who are building datacenter and data warehousing infrastructure have the most
immediate need/want of rackmount systems. In addition, ISPs can utilize a large
number of 1U or 2U appliance servers in a single rack. In general, customers
looking to locate a large number of processors (or processors and drives) in
a central spot will gravitate to racks and rackmount equipment.
rackmount systems are not recommended for small groups with modest computing
needs, e.g. workgroup/small department-only computing environments.
particular vendors: because of the consolidation underway, users may want to
be cautious about choosing one of the smaller players in the market.
24x7: 24 hours a day, 7 days a week
U: "rack units" - 1.75" vertical increments used to size rackmount
equipment. (For example, 4U = 7" high system) 4P" four-processor server
(similarly: 1P, 2P, 8P)
VAR: Value Added Reseller (a middleman between manufacturer and customer)
ISP: Internet Service Provider
RM: Rackmount or rackmountable
CAGR: Compound Annual Growth Rate