Product
Background
The product being reviewed is Hewlett-Packard's new six-CPU Intel-based
rackmount server, the NetServer LT 6000r. This product was released in
March, 2000, its purpose to raise the bar on CPU density for larger rackable
systems. This product is an enterprise-class machine, geared toward larger
computing environments. (Hewlett-Packard also released their NetServer
LH 6000 at the same time. The LH 6000 is also a six-CPU server, but is
more focused on internal disk expansion than on packing a bunch of CPUs
into a rack.)
The
NetServer LT 6000r is aimed at those users who want to increase the number
of "computes per square foot" of floor space. As floor space becomes more
expensive, customers tend to put units into equipment racks, expanding
upward instead of outward. Clearly, the more computing power that a customer
can pack into a given footprint, the greater the value.
Prior
to the LT 6000r's release, the industry norm had been four CPUs in a 4U-high
enclosure (1U = one rack unit = 1.75"). HP's offering increased the density
by 50%, which in turn increased the amount of computing power available
in any given rack space. For example, Compaq's ProLiant 6400R (now renamed
the DL580) can provide 40 CPUs in a 42U rack, the LT 6000r can provide
60 CPUs. It's other 4U competitors include the Dell PowerEdge 6350/6450,
and IBM's Netfinity 6000R.
This
product is a big step for HP, whose servers - with the exception of the
NetServer LPr - had generally been not much different from the rest of
the "Big Four" Intel server manufacturers (Compaq, Dell, HP, IBM). Although
the IBM Netfinity 4000R and the Compaq ProLiant DL360 can provide more
total CPUs in a 42U rack - 84 CPUs - those products are aimed more at
ISPs and ASPs, and less toward enterprise computing.
As
with the server market in general, this market will continue to grow,
although a little slower than the market for very thin (1U and 2U) servers,
and slower than server appliances. We expect the six-CPU product to take
away some market share from the 4U/4-CPU market.
Product
Strategy and Trajectory
Hewlett-Packard is trying to position their six-way (six CPU) servers
as high-density, high-performance, low-cost machines. In other words,
more bang for the buck, and more bang period. Until these products, HP
was generally in third or fourth place of the major Intel server manufacturers
(out of four) for price/performance. These products have changed that
drastically, vaulting HP into first place (albeit for only a short time).
Strategy
is one thing, trajectory is something else. Hewlett-Packard's NetServers
are conspicuous by their decreasing market share. Although HP sold more
NetServers in 1999 than in 1998, overall market share percentage has declined
for them - they are now in fourth place for the "Big Four" Intel server
vendors (Compaq, Dell, HP, IBM). One of our colleagues has called them
"the quiet company" relative to market presence. Even more telling is
that the other three majors don't even consider them a credible market
share threat any more, as evidenced by a number of competitive positioning
statements. Clearly, HP needs to regain lost momentum if it wishes to
succeed against the other three. A potential positive for HP: they believe
they are capturing a significant portion of the 4-CPU market (at which
the 6000 series is aimed) - but figures for that segment are difficult
to verify.
Product
Strengths
Performance
In terms of general transaction processing, the LT 6000r performance numbers
are among the highest, on the TPC-C benchmark scale, of any 4U-high server.
This is a strong selling point for those customers who want to pack as
much transaction processing capability into a given space. (This includes
things like database transactions.)
Price/performance
HP had traditionally been a laggard in price/performance, as measured
by $/tpmC. Its figures were typically several dollars per tpmC worse than
competing offerings from Dell and Compaq (where $15-$18 was considered
a good range), and only IBM's Netfinity servers (of the Big Four) cost
more per unit transaction. With the LH 6000 (the LT's "sister" server),
HP blew past the competition, racking up an impressive $13.95/tpmC. Although
this number was later bettered by a Compaq 8 CPU server figure, it shows
a renewed interest in providing cost-effective performance. We have discussed
HP's relatively poor price/performance in the past; we are glad to see
that they appear to be changing their focus in this regard.
System
density
As mentioned earlier, six CPUs in a 4U box challenges Compaq, Dell, and
IBM to match it. The performance advantages were obvious from viewing
the benchmark scores. It is also clear that Compaq and Dell have gotten
the message, as they have recently retested their 4U servers in an attempt
to challenge the 6000's numbers.
Product
Challenges
Most of the challenges associated with this server are company-related,
not product-related.
Linux
strategy
HP still has not convinced many people that it has a Linux strategy. Although
they have said some of the right words, we do not believe they have a
fully-integrated plan, whereas Dell, IBM, and Compaq (HP's primary competition
in the Intel server market) seem to be able to articulate and implement
a more cohesive Linux strategy.
Product
ordering/sales
Although HP has improved their online purchasing process, they still have
a way to go. Currently, less than 50% of their product line is available
through Web ordering, as compared to 100% for Dell, and 80% for Compaq.
A six CPU system is beyond the "small server" range, but it is still not
the beefiest server HP ships, and a 4U-high unit is (physically) small
enough that direct sales should be possible.
CPU/System
Serviceability(LT6000r)
Although the means of access to service the CPUs is interesting and simple
to open, we are not sure that it is as effective in practice as in theory.
Inserting and removing CPUs from a cantilevered surface (which is what
the flipped-up sheet metal becomes) is not as effective as pushing against
a fixed surface. In addition, the operational position works against gravity,
not with it ("hanging from" vs. "lying on"). For the PCI modules, we would
have preferred top access, not side access. In addition, it is difficult/impossible
to replace the "bottom" PCI card when the unit is installed in the rack
(yes, when the slides are extended). Even though that particular slot
is not hot-pluggable, having complete access would be preferable. (This
may be an impossibility, given that six slots requires nearly six inches
out of the seven-inch enclosure height.
We
feel that these are relatively minor considerations, given that HP has
managed to pack so much into the chassis.
Relative
performance
Although the TPC performance set a new standard when the results were
first released in March, Compaq and Dell have recently released four-CPU
performance figures that put them in the ballpark, or even surpass, the
6000's numbers. This means they can produce the same transaction-processing
performance with two-thirds of the CPUs. Dell and Compaq accomplished
this using 700 MHz processors, vs. the 550 MHz processors used by HP,
so the comparison is not truly apples-to-apples. However, unless HP can
match the numbers of Dell and Compaq, they run the risk of users deciding
that having six CPUs is more sizzle than steak.
BOTTOM
LINE
Vendor Recommendations
As
mentioned above, the six-way servers are relatively inexpensive, strong
performers. From a product standpoint, except for the modest serviceability
issues described, HP has produced a good design. They need to continue
in that manner. They also need to respond strongly (product-wise, not
spin-wise) to Compaq's and Dell's four-CPU performance improvements.
The
areas where HP needs to pay more attention are in the corporate realm.
Until recently, HP's website was rarely updated, and generally not nearly
as useful as the competition's sites. In addition to the lack of maintenance,
the site tended to present information that was questionable (refer to
TEC Note: "Intel
Small Server Market"). A current example is the performance section
talking about the "industry standard SPECweb96 benchmark" performance.
However, when they tested the LT to SPECweb99 (the current benchmark,
since web96 was retired in April, 2000), its performance was good, but
not great.
Fortunately,
someone at HP has decided to clean things up: in addition to a "Tell us
what's right/wrong" survey, we have noticed the site is more up-to-date,
and has been cleaned up a little. However, we would still like to see
more performance figures - they're still pretty sparse.
HP
needs to articulate its Linux strategy in a more concrete (and bold) way.
As mentioned earlier. the market's current response seems to be "HP has
a Linux strategy? Since when?" Both Dell and IBM have taken strong stances
on Linux, and Compaq is not far behind. With Linux accounting for 25%-30%
of new server shipment OS installs, HP cannot afford to miss the boat.
In
summary, Hewlett-Packard needs to improve the various support structures
for the products.
User
Recommendations
As mentioned earlier, the LT and LH 6000 products are aimed at enterprise-class
computing needs. They are not geared toward such customers as small ISPs,
where a "bunch of thin servers" is more appropriate; nor should the 6000s
be considered where server appliances are needed.
The
six-CPU products are strong performers, and the LT 6000r provides a high
density rackmount solution for anyone needing lots of transaction processing
(TP) power. In a filled two-meter rack, the LT 6000r can provide anywhere
from 29% to 65% more transactions per minute, when compared to offerings
from Compaq [ProLiant 6400R and 8500], Dell [PowerEdge 8450], and IBM
[Netfinity 8500R]. Add to this the low $/tpmC cost, and we believe TP-intense
customers will do well with this machine.
The
caveat here is the recent performance improvements of the competition.
Although Compaq's servers had gained a reputation for lower quality, people
still buy them about 2.5 times as much as HP servers, and Compaq has committed
to improving their server quality across the board. HP's servers are considered
very reliable, but not appreciably more so than Dell. [Reliability comments
source: Computerworld survey, 11/99]. Naturally customers
need to factor in reliability, not just performance, when buying an enterprise-class
machine.
If
HP can improve their "relative performance" numbers (as described earlier),
then this will provide more pluses for customers considering the 4U box.
However, even without the performance improvement, the price/performance
figures are sufficiently good that debates over whether it took four or
six CPUs to attain a given performance level become almost a non-issue.
In
summary, the LT 6000r server is a strong product from Hewlett-Packard,
and should be included in a user's "short list" of four-CPU servers under
consideration.