What does Sun's top of the line look like?
Written: Dec 29 '00 (Updated Mar 04 '01)
|
Product Rating:
|
|
| Customer Support: |
 |
|
| Reliability: |
 |
|
| Ease of Opening Box: |
 |
|
|
Pros: service, configurability, performance, adaptability
Cons: cost, cpu upgrade, trade-in value
The Bottom Line: A great top-of-the-line Solaris platform with excellent configurability features for a dynamic environment.
|
|
|
| jmccorm's Full Review: Sun Enterprise 10000 |
Three years ago, our site was delievered one of the first Sun E10ks. Today, our site has ten of them, and I personally administrate two of these frames in a production environment. The E10k is an incredible piece of server technology. But there are pros and cons to be aware of.
When Sun first introduced the E10k, they believed it to be a niche item and the demand wouldn't exceed more than a dozen units. They couldn't have been more wrong. The scalability and features of the box were incredibly popular -- so much that some of its features have trickled down into the smaller x500 server lines and into Sun's yet to be released (as of this writing) high end Serengetti architecture.
Before I get into how we've deployed our E10ks and my experience with them, you'll need to understand some of the technical details. And they're plenty. (NOTE: Specs are as of 12/29/00.)
The E10k frame is capable of holding up to 16 system boards, with a minimum of 4 boards. Each system board can hold four CPUs (400mhz), 4GB RAM (4 banks of 1GB), and either 4 SBUS devices, or 2 PCI devices. [Note: in the future, this *may* support faster CPUs, 2x higher memory density, and 3 PCI devices per system board.]
The frame is administrated through an SSP (System Service Processor) which takes the form of either an E250 or an Ultra 5. The SSP controls the booting, error handling, and configuration of the frame. With the SSP, you configure the physical hardware into logical "domains".
The domain concept, similar to mainframe partitioning, is important to understand. Although the E10k is a single physical machine, it can be divided into 1 to 16 logical servers that are electronically isolated from each other. The division between domains are made at the system board level.
As a small example, if you have an E10k with four system boards, you could create three domains. The first system board could be made into a domain called "happy". The second system board could be made into a domain called "sad". The third and fourth system boards could be combined into a domain called "curious".
Each domain runs a completely isolated version of the Solaris operating system, and each domain can be treated as a completely different server. If "happy" has a panic, and the power is turned off to "confused", then "curious" will continue to run because it is electronically and logically isolated from the other servers.
I'm spending a lot of time on the domain concept, because it isn't easy for some people to grab. On the hardware side, the E10k has a crossbar bus that connects system boards together (or acts as the electronic "glue"). Boards that are connected together can make a single domain that is more or less the same as any other Sun server. Because the crossbar isolates system boards from each other, what happens on one system board does not affect other system boards.
In short, you've got a frame that can hold up to 16 system boards, and the glue to create anywhere from 1 to 16 *isolated* systems inside the frame.
The E10k introduced a very popular feature of the E10k known as "dynmical reconfiguration" or DR. In short, DR allows you to, while a system is running, to add or remove system boards to the domain.
Need to add more CPUs and memory to a system? You can do it while its running. Need to remove a system board to add some more memory? If you've done you're configuration correctly, you can remove a system board... yes, with the operating system still running, add the memory to the board, and place it back into the domain.
DR is probably one of the coolest features of the E10k, and has been ported to the Ex500 server line.
Why would someone want an E10k? There are a number of reasons. The main reasons I see are that it allows you to deal with an unpredictable environment where things are changing, and it provides a great degree of scalability. As well, Sun takes the support of the E10k *VERY VERY* seriously. Expect a very high degree of support from the initial configuration of the system to the day-to-day support issues.
There are a many ways an E10k can be configured and used. But I'll focus on how our E10ks are configured. Generally, we dedicate an E10k as either production or test/development. However, it is just as possible to mix production and test within the same frame.
A domain can be treated just like any other Sun server out there. You can hook it up to Sun storage, or a third party. We prefer to use Sun Unipack disks for the boot devices, and EMC storage arrays for the applications.
Sun offers InterDomain Networks or IDN. Its a very good idea, but we've decided to avoid it. IDN creates a connection between two domains that were previously isolated. The problem is that when you have two domains connected via IDN, a panic in one domain has the possibility of taking down the other domain. The speed is good, but we've decided that the risk isn't worth the benefit.
The Dynamic Reconfiguration feature has pulled us out of the fire several times. You've suddenly got an excess demand (like memory or CPU) that your domain can't handle. What do you do? Add in another system board on the fly (again, while the OS continues to run) and it'll immediately take advantage of the additional resources. An absolute life saver, especially if you are up against a service level agreement. Just make sure your patch levels are up to rev.
Probably not a standard configuration, but in our production environments, we order an extra system board and call it a "spinner". If we have to add resources on the fly, we can do it. And we've got the on-site hardware in case of a system board failure. It is a practice which I wish Sun would recommend.
The E10k currently support 250, 336, and 400mhz processors. Of course, any new E10k should be purchased with the 400mhz processors. But there are a few things to be aware of here.
First, if you need to upgrade the CPUs in the E10k, you must upgrade all the CPUs in the *entire frame* at once. You just can't upgrade one domain. You're looking at a fairly hefty cost and some down-time for this operation.
I'm personally a little disappointed in the Sun CPUs. We almost certainly can't expect more than 600mhz to be released in the current generation of hardware. But the current CPUs are still quite powerful.
Finally, there is the CPU external cache reliability issue, but any E10k purchased from this point forward should be receiving the mirrored L2 cache, which completely addresses this issue above and beyond the current kernel patches.
On Sun's web page, they've placed a lot of emphasis on the bandwidth of their "Gigaplane-XB Interconnect". I can't say that I've ever had a problem that I've been able to point to the bus at, but then again, there really aren't the tools available to see this. I'm told that the E450 has the least bus latency, with the E10000 having the most.
As mentioned earlier, the level of service from Sun for the E10k is incredible. A Sun engineer will help you with the initial configuration, and make sure that it is "blessed" by Sun. The telephone support is top-notch for the E10k. It comes with a default "platinum" contract. Sun insists that only their engineers touch the internal E10k hardware and system boards -- not a bad idea.
All and all, I consider this to be an incredible system. As a systems administrator, I'd far prefer to administrate an E10k platform over anything else.
If you're considering purchasing an E10k, the best thing to do is to talk to several customers and get a feel from several customers (and perhaps at the management and technical levels) and see what they have to say beyond Sun's web site specifications.
UPDATE: March 4th, 2001
Shortly after this review was created, it was (almost verbatim), placed on another website, and then featured as a Slashdot headline under "Million Dollar Reviews". It took a bit of work, but I was given credit as this being my original work.
One important benefit that I neglected to point out was the "netcon" feature which allows remote console access to a domain from the SSP. Over a network connection, you can have direct console access to any Sun E10k, similar to the way the RSC works on an E250. An absolute lifesaver for remote administration.
Also, another product has come to my attention that I, as of yet, do not have a great deal of information about. Fujitsu has a Sun compatible prduct line (running Solaris). The high end models appear to be better than the E10k (up to 128 processors at 500+mhz). Allegedly has good service, through Amdahl, and a very attractive price. It is worth checking out.
Recommended:
Yes
|
|
|
|
Epinions.com ID: jmccorm
|
|
Member: Josh McCormick
Reviews written: 16
Trusted by: 5 members
|
|
|