Scalable NAS
Over the past three years, humanity has produced more information than in its entire history until 2010. The accumulation of data continues exponentially and no one is surprised by the customers' requirements to store petabytes of data. The use of traditional arrays in such cases is often irrational, and sometimes impossible. Of course, high-level storage systems using low-performance SATA disks will be able to provide the customer with more petabyte data, but this solution will be expensive and slow. To speed up the processing of relevant amounts of information the most optimal use of parallel processing. Along this path designers of computer systems have long been following. If you take a list of the most productive systems - it will be a cluster of hundreds and thousands of small servers. The same thing happens in storage systems - providing acceptable I / O with volumes of more than a petabyte can only be achieved using a cluster of storage systems. Therefore, it is not surprising that almost all equipment manufacturers have in their product line scalable NAS storage systems (scale-out NAS).
The article made an independent review of existing scalable NAS.
The table shows the comparison of the characteristics of Scale-NAS systems. Performance results are available at www.spec.org/sfs2008/results/sfs2008nfs.html
')
Gluster
The gluster project is based on free software. The advantage of this approach is the lack of binding to any specific storage systems. In addition, thousands of programmers involved in a project can be involved in solving a software crash problem, which ultimately leads to a faster solution than commercial systems.
Due to the distribution of files across servers using a specific algorithm, the cluster of NAS servers does not require a dedicated server to store metadata.
Gluster has no practical limit on the number of nodes, supported data volume and is currently the most scalable NAS solution.
Red hat storage server
Red hat storage server is based on open source development (the gluster project). Licenses for Red hat storage server, as well as for gluster, are free. The customer pays only for support. Due to the fact that Red hat assumes certain obligations for support, the customer can only deliver the equipment configurations tested by Rad hat.
It is worth noting that in the case of Red hat storage server the advantage of using open source code is preserved. If the software fails, other programmers involved in the Gluster project may be involved in solving the problem.
Red hat storage server supports up to 64 nodes with a total disk space of up to 8 pb.
Netapp Clustere Data Ontap
Netapp was one of the first to offer NAS solutions and, of course, it could not bypass the scalable NAS. FAS / V 62XX series devices can be combined into a NAS cluster with a common name space consisting of 24 devices.
The maximum amount of cluster disk space is 69 PB, which is a record among commercial systems. The capacity of a single volume (infinite volume) can be 20 PB and at the same time use up to 10 Netapp storage systems.
The test results (http://www.spec.org/sfs2008/results/sfs2008nfs.html) showed good performance. Moreover, when comparing the results in 16, 20 and 24x of the node configuration, a linear increase in productivity is seen with an increase in the number of cluster nodes.
EMC Isilon
EMC decided not to develop its own product, but to purchase a ready-made solution for scalable NAS. This solution was the company Isilon systems, now offered as EMC Isilon. The solution is scaled to 144 nodes and supports up to 15 pb. Of the advantages of the system is the ability to distribute one file across several nodes and set a storage policy for each file. The reverse side of this advantage is that the information on the location of individual parts of the stored files requires an increased size of metadata, the further processing of which reduces system performance. Indeed, even in the maximum configuration with 140 nodes of the cluster, EMC Isilon shows mediocre results both in the number of requests processed and the processing time (http://www.spec.org/sfs2008/results/sfs2008nfs.html).
Huawei
According to performance tests (http://www.spec.org/sfs2008/results/sfs2008nfs.html), Huawei on a cluster of 24 NAS systems N8500 shows a record for scalable NAS systems 3064602 ops / sec performance. This is three times better than the Isilon 140 node cluster.
In the maximum configuration, the Huawei cluster supports up to 24 nodes with a total capacity of 15 PB. In addition, the NAS provides the ability to dynamically redistribute the location of files depending on the load DST (dynamic storage tiering). In addition to record-breaking performance, in the absence of a load on the disks, the system disables their power consumption (drive speed down function), which is very important for storage systems with disk volumes from petabyte and higher.
Conclusion
Almost every manufacturer of IT equipment has scalable NAS solutions in its portfolio. In addition to the above, scalable NAS solutions are available from HP, IBM and others.
IBM offers IBM Scale Out Network Attached Storage (SONAS) with support for up to 60 nodes in a cluster and 28 pb data. HP offers HP Irbix with the ability to scale up to 16 petabytes and 1024 nodes in a cluster. We will not consider these solutions in detail, as they do not have outstanding performance characteristics, supported space, etc.
Consideration of the listed systems shows that, depending on customer requirements, the choice of the most optimal NAS cluster may be different. Of course, the most budget option will be when using Gluster. In addition to the lack of payment for software, the use of existing storage systems in the company will allow to save previously made investments and reduce the cost of purchasing new equipment. The use of Red hat storage server limits the use of existing equipment, but at the same time provides guaranteed support from the vendor. It makes sense to consider using EMC Isilon when working with large files when it is necessary to distribute one file among several cluster nodes. If you need maximum performance - then this is Huawei. If the customer is prohibited by law to use Huawei (for example, restrictions are imposed in the United States), then the most optimal choice would be a scalable NAS cluster from Netapp.
The article made an independent review of existing scalable NAS.
The table shows the comparison of the characteristics of Scale-NAS systems. Performance results are available at www.spec.org/sfs2008/results/sfs2008nfs.html
')
Gluster
The gluster project is based on free software. The advantage of this approach is the lack of binding to any specific storage systems. In addition, thousands of programmers involved in a project can be involved in solving a software crash problem, which ultimately leads to a faster solution than commercial systems.
Due to the distribution of files across servers using a specific algorithm, the cluster of NAS servers does not require a dedicated server to store metadata.
Gluster has no practical limit on the number of nodes, supported data volume and is currently the most scalable NAS solution.
Red hat storage server
Red hat storage server is based on open source development (the gluster project). Licenses for Red hat storage server, as well as for gluster, are free. The customer pays only for support. Due to the fact that Red hat assumes certain obligations for support, the customer can only deliver the equipment configurations tested by Rad hat.
It is worth noting that in the case of Red hat storage server the advantage of using open source code is preserved. If the software fails, other programmers involved in the Gluster project may be involved in solving the problem.
Red hat storage server supports up to 64 nodes with a total disk space of up to 8 pb.
Netapp Clustere Data Ontap
Netapp was one of the first to offer NAS solutions and, of course, it could not bypass the scalable NAS. FAS / V 62XX series devices can be combined into a NAS cluster with a common name space consisting of 24 devices.
The maximum amount of cluster disk space is 69 PB, which is a record among commercial systems. The capacity of a single volume (infinite volume) can be 20 PB and at the same time use up to 10 Netapp storage systems.
The test results (http://www.spec.org/sfs2008/results/sfs2008nfs.html) showed good performance. Moreover, when comparing the results in 16, 20 and 24x of the node configuration, a linear increase in productivity is seen with an increase in the number of cluster nodes.
EMC Isilon
EMC decided not to develop its own product, but to purchase a ready-made solution for scalable NAS. This solution was the company Isilon systems, now offered as EMC Isilon. The solution is scaled to 144 nodes and supports up to 15 pb. Of the advantages of the system is the ability to distribute one file across several nodes and set a storage policy for each file. The reverse side of this advantage is that the information on the location of individual parts of the stored files requires an increased size of metadata, the further processing of which reduces system performance. Indeed, even in the maximum configuration with 140 nodes of the cluster, EMC Isilon shows mediocre results both in the number of requests processed and the processing time (http://www.spec.org/sfs2008/results/sfs2008nfs.html).
Huawei
According to performance tests (http://www.spec.org/sfs2008/results/sfs2008nfs.html), Huawei on a cluster of 24 NAS systems N8500 shows a record for scalable NAS systems 3064602 ops / sec performance. This is three times better than the Isilon 140 node cluster.
In the maximum configuration, the Huawei cluster supports up to 24 nodes with a total capacity of 15 PB. In addition, the NAS provides the ability to dynamically redistribute the location of files depending on the load DST (dynamic storage tiering). In addition to record-breaking performance, in the absence of a load on the disks, the system disables their power consumption (drive speed down function), which is very important for storage systems with disk volumes from petabyte and higher.
Conclusion
Almost every manufacturer of IT equipment has scalable NAS solutions in its portfolio. In addition to the above, scalable NAS solutions are available from HP, IBM and others.
IBM offers IBM Scale Out Network Attached Storage (SONAS) with support for up to 60 nodes in a cluster and 28 pb data. HP offers HP Irbix with the ability to scale up to 16 petabytes and 1024 nodes in a cluster. We will not consider these solutions in detail, as they do not have outstanding performance characteristics, supported space, etc.
Consideration of the listed systems shows that, depending on customer requirements, the choice of the most optimal NAS cluster may be different. Of course, the most budget option will be when using Gluster. In addition to the lack of payment for software, the use of existing storage systems in the company will allow to save previously made investments and reduce the cost of purchasing new equipment. The use of Red hat storage server limits the use of existing equipment, but at the same time provides guaranteed support from the vendor. It makes sense to consider using EMC Isilon when working with large files when it is necessary to distribute one file among several cluster nodes. If you need maximum performance - then this is Huawei. If the customer is prohibited by law to use Huawei (for example, restrictions are imposed in the United States), then the most optimal choice would be a scalable NAS cluster from Netapp.
Source: https://habr.com/ru/post/201698/
All Articles