Storage Dual-Live configuration considerations across data centers from an Oracle RAC perspective

ORACLE RAC was designed to be dual-live across data centers, designed to load balance and high availability for multiple hosts with shared storage in a data center. But since its architecture does have the potential to load balance and high availability across data centers, several storage vendors have expanded its use environment and presented solutions across data centers. ORACLE The default attitude is taken, but it is recommended that all solutions be carefully tested before they are put into production by the customer.

forRAC, the biggest bottleneck across data center solutions is between nodesInterconnectbecause of its very high latency and bandwidth requirements. Generally speaking, localInterconnectTransmission delay in1~2msbetween, localIOthe delay is in the8~15msbetween. These two delays have a significant impact on performance, and if you use a dual data center scenario, they can seriously affect performance as the room distances grow. And becauseInterconnectThe latency base is low (1~2ms), resulting in a delay in the machine room distance to the entireInterconnectthe proportion of the impact is greater: think about if the distance is prolonged2msthe transmission delay, forInterconnectis that100%~200%of delayed growth, forIOthen only15%~25%of growth. Of course, withSSDlarge amount of use in storage, distance toIOThe impact is also increasing.

to visually demonstrate the transmission distance IO and the Interconnect the effect of the delay, figure one and Figure II show the HP test results as a reference:

650) this.width=650; "Title=" from the Oracle RAC Perspective storage Dual-live configuration considerations for cross-Datacenter 1.png "style=" Float:none; "src=" http://s3.51cto.com/ Wyfs02/m00/82/10/wkiol1dkzvztdabwaabbh6np5u8130.png-wh_500x0-wm_3-wmp_4-s_3921625582.png "alt=" Wkiol1dkzvztdabwaabbh6np5u8130.png-wh_50 "/>

Figure A

figure one shows the io buffer-to-buffer Credits (BBC) BBC Feature allows a large number of unanswered packets to be kept in the cache while the packets continue to be sent. In the case of large data traffic, the farther away, the BBC

if at distance 100km BBC io latency is approximately increased from local 43% Span style= "font-family: ' The song Body '; If you do not open BBC io delay approximately increase 120~140% 20km distance, not open BBC 20~24%

Figure Two is the use of high-load and low-load pair configuration one or two Interconnect of the RAC tested and examined the distance Interconnect the impact.

650) this.width=650; "Title=" from the Oracle RAC Perspective storage Dual-live configuration considerations for cross-Datacenter 2.png "style=" Float:none; "src=" http://s3.51cto.com/ Wyfs02/m00/82/10/wkiol1dkzv3j-6m2aaeeyq6dgm4413.png-wh_500x0-wm_3-wmp_4-s_1249710467.png "alt=" Wkiol1dkzv3j-6m2aaeeyq6dgm4413.png-wh_50 "/>

Figure II

Two findings of this test are shown in Figure two:

1. Two links can reduce 50% latency in high load situations compared to one link

2. 100km can bring about 1ms of delay increase.

figures I and II show the effect of distance on the link, the following might and figure four show distance RAC the impact of overall performance.

in the long-distance transmission process, buffer-to-buffer Credits (BBC) function has a great impact on transport performance, so it is important to highlight figure III showing two manufacturers in open BBC the test results obtained in the case of the function. In contrast, figure IV shows the Results of the test without opening the BBC's function.

650) this.width=650; "Title=" from the Oracle RAC Perspective storage Dual-live configuration considerations for cross-Datacenter 3.png "style=" Float:none; "src=" http://s3.51cto.com/ Wyfs02/m01/82/12/wkiom1dkzgogz7knaabkgzbynjy832.png-wh_500x0-wm_3-wmp_4-s_3262459784.png "alt=" Wkiom1dkzgogz7knaabkgzbynjy832.png-wh_50 "/>

650) this.width=650; "Title=" from the Oracle RAC Perspective storage Dual-live configuration considerations for cross-Datacenter 4.png "style=" Float:none; "src=" http://s3.51cto.com/ Wyfs02/m01/82/12/wkiom1dkzgtjog6jaabs8gry6ry599.png-wh_500x0-wm_3-wmp_4-s_2671369827.png "alt=" Wkiom1dkzgtjog6jaabs8gry6ry599.png-wh_50 "/>

as you can see from figure III and figure IV, open BBC , the performance of the two test vendors in the scenario is pretty good. But if the BBC is not turned on, performance will fall sharply as distance increases. Considering that the same machine room configuration is better than the two-node RAC performance is about the single-node high 30~60%, if because the remote room RAC if the cluster has a performance degradation greater than 20% , consider using The RAC scenario carefully.

There are two points to note:

1. The test results given by the manufacturers are often the best data measured in the case of extreme optimization, the actual customer site optimization is often significantly lower than the factory test environment

2. manufacturers often only give their own optimal test results. The range of tests given by the two manufacturers in the three is not the same, due to the possibility of exceeding the range, the performance will have a large decline.

based on the above test, ORACLE It is recommended to consider whether to use the distance of the cable from the connecting room RAC Dual Live Solutions:

1. A room with a distance of less than 50km may consider using a dual-live RAC.

2. distance greater than 50km, less than 100km Room, carefully consider the use of double live rac< /c14>. To be used, a very careful test is required.

3. The distance is greater than 100km, it is not recommended to use a dual live RAC, you can consider RAC one node c8> do high reliable cluster ①.

① RAC one node is a RAC A variant of the effect somewhat similar to the traditional HP mc/sg + Oracle scenario, because at the same time only one node is running, there will not be a lot of data running in Interconnect on.

If you decide to use a cross- datacenter RAC , the following configuration recommendations need to be considered carefully:

1. interconnect and IO Links use non-shared, end-to line cable direct connection, English called "Dark Fibre" .

2. It is highly recommended to turn on the BBC function on the transmission path .

3.        in theORACLE Clustwarein the configuration3avoting DiskorVoting file. Each of the two data centers is equipped with onevoting Disk, in addition to a third room configuration based onNFSorISCSIof theVoting fileto improveRACsystem reliability.

With the previous test results, the first two points are easier to understand, below we have a detailed description of the 3rd recommendation:

If you do not configure a third room-based Voting file , when the two data room links are broken, both sides of the host can only access local storage, and do not know the other side status. At this point, because there is no third-party quorum, the RAC hosts on both sides exit the cluster, causing a business outage. Because if this is not the case, it will result in data disruption and more serious consequences.

Remote Voting file Configuration Considerations for:

Generally speaking,Oracle Clustwareper second by reading and writing less than1gigabytes of data accessVoting fileone time. Each write requestIOthe response should be in $seconds (default,Long Disk timeout) or -in seconds (configurable,Short Disk Timeout) returns. To this end,OracleRecommendedvoting Fielthe WriteIOshould be in -(27/2) Within seconds, the transmission bandwidth is at least128k bps.

Storage Dual Live and RAC the arbitration competition problem of the cluster

l   hp XP7 voting disk/file Configure to a virtual disk array, can avoid competition. Because there is no access to the rac nodes are not likely to be rac clusterware arbitration for the living. In this case, rac configure remote voting file

l   hp 3par this use alua The quasi-storage dual-live scheme for the protocol, because rac Span style= "font-family: ' The song Body '; > The node only uses one physical array at a time, with the result of xp7 Similarly, just put voting disk all configured as peer persistence rac configure remote voting file

l for other storage-dual-live solution providers that do not use Virtual disk array technology, especially for local read-write optimization, this is a very serious issue to consider. Since most of the arbitration for this storage-dual-live solution provider is implemented using a third-location virtual machine, it is recommended that the virtual machine be physically close to the third voting file of the RAC as far as possible. Reduce the likelihood that differences in physical factors result in a conflict of arbitration results.

l Some storage providers provide the means to manually adjust the quorum algorithm to ensure that the result of the storage quorum RAC same. Because there is no detailed information, so it is inconvenient to comment, but Oracle officials are opposed to this.

Bibliography:

the Oracle RAC and Oracle RAC one Node on Extended Distance (stretched) Clusters "

the Using Standard NFS-to-support a third voting file for extended cluster Configurations-oracleclusterware 11g Release 2"

the Oracle Clusterware Administration and Deployment Guide "

the HP 3Par Remote Copy software User ' s guide "

This article is from the "splinters nonsense" blog, make sure to keep this source http://leesbing.blog.51cto.com/1344594/1784197

Storage Dual-Live configuration considerations across data centers from an Oracle RAC perspective