In the past two years, the hottest thing in the server field is the blade server. It seems that it can replace all other server architectures. But the author believes that you should not mislead all kinds of advertising. Blade servers are not omnipotent and do not have an absolute advantage. Nowadays, people usually like to compare the blade server with the density of the rack server. This section will introduce some experience in the selection of the two.
in comparisons blade servers and rack servers, our main concern is the product of the investment cost, performance, configuration, scalability, energy consumption, maintenance and management. The following is a comparison of specific examples. This example is IBM's HS21 and HS21 XM blade servers, and the equivalent x3550, x3650 rack server.
HS21 and HS21 XM two blades are IBM Blade Center H (BC-H) for the chassis. The BC-H chassis consists of a 10GB Ethernet switch, a 1GB Ethernet switch and four power supply systems. The XM series blades are equipped with a pair of 2.33GHz Intel quad-core processors, while the HS21's pair of quad-core CPUs have a frequency of 2.0GHz; the XM's memory is 16GB DDR2, while the HS21 has only 8GB.
above configuration and IBM x3550 rack servers difference is not big, but clock speed a bit, as 2.66GHz. The IBM x3650 is a 2U dual quad-core rack server with a processor frequency of up to 3.0GHz.
l investment costs are
we all know, expensive blade servers, and then talk about the total cost of investment we are most concerned about. First look at the comparison of single machines (prices from IBM online quotes, are standard).
The standard HS21 blade server is priced at $7,100, while the HS21 XM blade server is priced at $9,800. The difference between the two is CPU frequency, memory size, and RAID 0/1 configuration on XM. The BC-H chassis used to mount the blades is up to $17,000. The rack server x3550 is priced at $7,900 and the corresponding x3650 is priced at $9,300.
A simple comparison shows that the blade server price for a single blade ($7,100+$17,000=$24,000, or $9,800+$17,000=$26,800) is lower than the single rack server due to the chassis price of the blade server. The price ($7,900, or $9,300) is much more expensive.
Imagine if HS21 or HS21 XM blade server in the two blades installed, the blade server price:
$ 7,100 * 2 + $ 17,000 = $ 31,000, or $ 9,800 + $ 17,000 = 2 * $36,600, and the price of two x3550, x3650 rack servers: $7,900*2=$15.800, or $9,300*2=$18.600. As a result, the overall cost of the blade server is still more expensive than the rack. But this does not mean that the blade server price is more expensive than the rack server, because we can see that in terms of a single "blade", it is indeed cheaper than a single rack server, just the blade chassis used. It dragged down the entire cost. As long as there is a proper "blade" size, the ultimate overall cost of the blade server will certainly be less than the overall low cost of multiple rack servers. And we have not considered the price of the switches required for multiple rack server connections.
If 14 "blades" are installed in the stencil chassis, the total cost of installing the HS21 "blade" is:
$7,100*14+$17,000=$116,400; if installed The HS21 XM "blade" costs $9,800*14+$17,000=$154,200.
The total cost of 14 x3550s is: $7,900*14=$110,600; the total cost of 14 x3650s is: $9,300*14=$130,200. Note that the final total cost also calculates the cost of a 24-port 1G fiber-optic Ethernet switch, approximately $12,500. The total cost of the above two server solutions is: $123,100 and $142,700, respectively. Comparing the total cost of the two blade server solutions above, it can be known that the total cost of the blade servers of the 14 HS21 "blades" is lower than that of the two 14 rack server solutions. The total cost of a blade server with 14 HS21 XM "blades" is comparable to the total cost of a 14-machine x3650 rack server solution.
l Operational performance comparison
Tested by some professional testing organizations, LBMench3 is one of the most demanding tests on the server - processor fork+execve (main test data is in I/The transfer speed between O and memory), the H21 and H.21 XM blade server test results are 289.9 microseconds, while the x.3550 rack server is 173.2 microseconds, the speed is increased by 40% (the longer the time, the speed The lower). This is due to the fact that the H.21 and H.21 XM blade servers are configured with RAID1, while the rack-mounted x.3550 and x.3650 servers are configured with RAID5. Other test structures show that the performance of these servers is comparable, because the server configuration of these servers is basically the same, they use two quad-core Intel processors, and they have chosen almost the same when testing on the main frequency. Configuration. In this way, there is not much difference between the blade server and the rack server in terms of performance. The main difference is only reflected in the data access capability affected by the disk array.
l scalability compare
on the expansion of capacity, we need to compare the two. One is the comparison of a single server, and the other is a comparison of server cluster extensions.
blade servers in a single server that each "blade." These "blades" have very limited expansion capabilities due to their high density. This is both an advantage of the blade server and an inevitable disadvantage. The general "blade" is only equipped with two hard disks, at most only RAID 1 mirroring, or RAID 0 has no redundant performance hard disk array, which means that the hard disk capacity can not exceed the maximum capacity specified by the blade manufacturer, and can not be maximized. Server disk read and write performance. This is why the processor fork+execve test results of the blade server we introduced earlier are far less than the rack server. Unless the server is installed and configured with iSCSI, Fibre Channel, or other SAN connectivity, it is difficult for the "blade" to substantially improve disk capacity and read and write performance.
blade server in the "Blade" different room for expansion rack servers to many large, which, like it can be configured more components. For example, the 2U x3650 rack server has built-in RAID 5 and is equipped with four hard disks: three for RAID and one for hot backup. Some manufacturers even placed a storage capacity equivalent to 2U in a 1U rack server.
addition to the expanded disk capacity aspects, memory, I /O expansion slots aspect, there is the same phenomenon that blade servers "Blade" scalability as good as rack-mounted servers.
Further, there is another potential disadvantage blade, it is proprietary. If you choose a vendor's blade server, it means that the service project and component updates must be followed by the vendor, and the services and components of other vendors cannot be used. Moreover, in general, future updates, faster CPUs will still be used first on the rack server.
scalability of the cluster, blade servers has inherent advantages, just put each "blade" into the blade chassis, using its own cluster management software can easily configure the blade server farms. Rack servers are not so easy to set up a cluster. It must purchase cluster management software separately, and the network connection and configuration are more complicated. Compare
l energy consumption compared to the performance, the maximum cost of the server is running - energy costs are often overlooked. Because rack-mounted servers have more components installed, and today's blade servers have optimized processor choices, blade servers generally consume far less power than rack-based servers. According to tests conducted by some testing organizations, the same configuration of blade server clusters and rack server clusters consumes nearly 20% less energy. This is not much different for small companies with only a few servers, but if there are dozens or hundreds of such servers in a company, the benefit is significantly increased. Compare
l management
in some large companies, server maintenance and management costs are also very high. In general, blade server maintenance is more convenient than rack servers. First, the size and weight of the blade is easy to handle. We can take a blade in 3 seconds, compared to pulling out a rack part for more than a minute. Moreover, the number of cables in the blade server (including power lines, twisted pair or fiber optic power bills) is much smaller than that of a rack server of the same size, because the "blades" share the same power supply, fan, The "blade" exchange base is provided by the blade chassis. The reduction of cables provides a lot of convenience for server maintenance and management.
In addition, as each blade in front of that blade servers can be managed through a unified cluster management software that comes with, and rack-mounted servers to be configured as a server cluster, you need to purchase additional management software group off, and The configuration is relatively more complicated.
But we also see that, because of technical blade in a relatively closed between manufacturers, blade server rack server can not do like that you can always replace each other, the lack of flexibility. This brings some inconvenience to the maintenance and management of the server, and may often require the service provider to support it. This must be fully realized.
view of the above comparison results in all aspects, we can see, rack-mounted servers and blade servers with no absolute advantage, which also determines the different business users have different options.
if just from the overall cost to consider, for just a few servers for small and medium businesses said the choice of rack-mounted servers more appropriate. Moreover, in terms of computing performance and scalability, rack-mounted servers are more secure and more suitable for the characteristics of fast-growing small and medium-sized enterprises.
if it is more than medium-sized enterprises, required more servers, the blade server is more appropriate choice. The management of the following aspects is easier, on the other hand, the overall cost and energy consumption can also save a lot of money. The lack of scalability of the blade server can be fully compensated by multiple "blade" clusters and SAN storage networks. These are almost standard configurations for medium-sized and above enterprises, and it can be said that performance equivalent to rack-mounted server sets can be achieved without additional investment.