Huawei Servers: Functional Testing
This article was prepared by Andrey Sidorov, Unix engineer of the Jet Infosystems Service Center.
Recently, one of our customers tested Huawei equipment, the results of which we would like to share with you. The main task of testing was to check the basic functions of Huawei equipment and their compatibility with the customer’s current infrastructure. At the "torn apart" we got a fairly large amount of iron - rack servers, blade baskets.
Servers for tests got the following:
')
The first two of them work with Intel Xeon E5- * V3 and DDR4, the second two - Xeon V2 and DDR3
Here we pulled out power supplies, fans, disks - on all servers, memory boards at 8100, completely cut off the power on each server. As a result of our actions, no server was damaged, they passed the reliability check, but several features were found.
Replacing the power supply and fans occurs regularly, according to the documentation. I was a little upset that at 1288 and 2288 the replacement of fans was only with the shutdown.
Replacing disks on any servers - if iron RAID is organized, then you need to install the utility from LSI on the host and only there to watch the rebuild process. Or reboot and look in the built-in utility. This information is not available from the management interface.
To hot-swap the memory-board on the 8100, you need to enable the hot-sparing for each row, then, if a certain number of errors are detected, the memory-board will move to the hot-spare board. The relocation mechanism can be either automatic or manual. Pros - replacing memory without shutting down the server, minuses - losing 1/8 of all memory in the system. The process of moving is long, 256GB - 5‒6 hours. It does not save if the board or memory strip suddenly dies, but for this there is a memory mirror. Pros: if all this is set up, the only thing that can knock down your system is a dead processor. Cons: losing 9 out of 16 memory boards.
We checked what information it gives, according to which protocols it can give it, according to which protocols the interface itself can be controlled. They, in fact, were 3:
HMM - basket management interface. Peculiar. You can configure the MM of the blades, but you cannot configure the gateway for switches via the web interface; you can use the CLI. It knows how to create profiles for slots with pre-defined MAC and WWN, but from the HMM it is not possible to switch to the blade web interface.
iMANA 200 - management interface for V2 generation servers and blades and some V3 models
iBMC is an interface for managing servers and blades of the V3 generation.
They have similar functionality, but differ greatly in design. Gentleman's set - you can see the status of the components, can send these statuses by mail or SNMP (can not syslog). Unfortunately, it is impossible to see the total amount of memory in the system, only by multiplying the number of planks by their capacity - there is such information. KVM is present, and it is completely normal. There is information about the current load of fans and power supplies. Bad friends with LDAP - spaces in the names of groups and OU, nested by OU are not supported. The domain controller server must be set as IP.
Available protocols for management are http (redirects to https), https, ssh, telnet (disabled by default). There is SNMP, V3 is enabled by default, it cannot be disabled. All information that is provided can be obtained via IPMI, SNMP. To upgrade firmware via CLI, there is ftp, turned off by default.
From the manufacturer's site, updates are available only for MM, BIOS. There are no firmwares for RAID controllers or HBAs. All firmware have to install one at a time, any boot image for a one-time update of all firmware Huawei has not yet done. Update iBMC, iMANA, HMM painless for OS.
They put VMware ESXi 5.5, RHEL 7.1, Win 2008R2 SP1, Win 2012 R2. There were no problems with the installation, the drivers are available on the Huawei website. The OS can be installed either as usual, or with the Huawei Service CD. In the latter case, the OS is installed with all the drivers at once.
Our testing has shown that, despite the shortcomings and comments found, there are no serious problems with reliability, there are no serious inconsistencies with generally accepted norms or customer standards.
Recently, one of our customers tested Huawei equipment, the results of which we would like to share with you. The main task of testing was to check the basic functions of Huawei equipment and their compatibility with the customer’s current infrastructure. At the "torn apart" we got a fairly large amount of iron - rack servers, blade baskets.
Servers for tests got the following:
- RH1288 V3 - 1 unit, 2 socket
- RH2288 V3 - 2 unit, 2 socket
- RH5885 V3 - 4 unit, 4 socket
- RH8100 V3 - 8 unit, 8 socket
')
The first two of them work with Intel Xeon E5- * V3 and DDR4, the second two - Xeon V2 and DDR3
| RH1288 |  |
|---|---|
| RH2288 |  |
| RH5885 |  |
| RH8100 |  |
| E9000 Blade Basket |  |
 |
Go through the points, as well as what we tested.
The first and one of the most important is reliability.
Here we pulled out power supplies, fans, disks - on all servers, memory boards at 8100, completely cut off the power on each server. As a result of our actions, no server was damaged, they passed the reliability check, but several features were found.
Replacing the power supply and fans occurs regularly, according to the documentation. I was a little upset that at 1288 and 2288 the replacement of fans was only with the shutdown.
Replacing disks on any servers - if iron RAID is organized, then you need to install the utility from LSI on the host and only there to watch the rebuild process. Or reboot and look in the built-in utility. This information is not available from the management interface.
To hot-swap the memory-board on the 8100, you need to enable the hot-sparing for each row, then, if a certain number of errors are detected, the memory-board will move to the hot-spare board. The relocation mechanism can be either automatic or manual. Pros - replacing memory without shutting down the server, minuses - losing 1/8 of all memory in the system. The process of moving is long, 256GB - 5‒6 hours. It does not save if the board or memory strip suddenly dies, but for this there is a memory mirror. Pros: if all this is set up, the only thing that can knock down your system is a dead processor. Cons: losing 9 out of 16 memory boards.
The second item of our program was testing the management interface.
We checked what information it gives, according to which protocols it can give it, according to which protocols the interface itself can be controlled. They, in fact, were 3:
HMM - basket management interface. Peculiar. You can configure the MM of the blades, but you cannot configure the gateway for switches via the web interface; you can use the CLI. It knows how to create profiles for slots with pre-defined MAC and WWN, but from the HMM it is not possible to switch to the blade web interface.
iMANA 200 - management interface for V2 generation servers and blades and some V3 models
iBMC is an interface for managing servers and blades of the V3 generation.
They have similar functionality, but differ greatly in design. Gentleman's set - you can see the status of the components, can send these statuses by mail or SNMP (can not syslog). Unfortunately, it is impossible to see the total amount of memory in the system, only by multiplying the number of planks by their capacity - there is such information. KVM is present, and it is completely normal. There is information about the current load of fans and power supplies. Bad friends with LDAP - spaces in the names of groups and OU, nested by OU are not supported. The domain controller server must be set as IP.
Available protocols for management are http (redirects to https), https, ssh, telnet (disabled by default). There is SNMP, V3 is enabled by default, it cannot be disabled. All information that is provided can be obtained via IPMI, SNMP. To upgrade firmware via CLI, there is ftp, turned off by default.
The third point is the firmware update.
From the manufacturer's site, updates are available only for MM, BIOS. There are no firmwares for RAID controllers or HBAs. All firmware have to install one at a time, any boot image for a one-time update of all firmware Huawei has not yet done. Update iBMC, iMANA, HMM painless for OS.
The fourth point was the installation of various operating systems.
They put VMware ESXi 5.5, RHEL 7.1, Win 2008R2 SP1, Win 2012 R2. There were no problems with the installation, the drivers are available on the Huawei website. The OS can be installed either as usual, or with the Huawei Service CD. In the latter case, the OS is installed with all the drivers at once.
Conclusion
Our testing has shown that, despite the shortcomings and comments found, there are no serious problems with reliability, there are no serious inconsistencies with generally accepted norms or customer standards.
Source: https://habr.com/ru/post/267421/
All Articles