 good afternoon everyone and and it's my honor to make such a such a sleek in this afternoon I will talk about the campaign comparison of safe and commercial solar sand this is our content we first we will introduce our company and then we will introduce the safe your and solid fire and the sky oh then we we will get your company or even all safe and solar sand and at the last start the speech we will give a performance test between safe and the sky oh first let me introduce Huawei our China our our company named AW Cloud and we are pure open-stack player in China and we are China's leading enterprise cloud service provider and we have broad deployment of production clouds in China and we have large-scale inflation operation in the background and highly diverse set of workloads and use case first let me introduce the safe oh everyone here should be very familiar with safe oh so I just give a shorter description of it safe is an open source object and it is delivered I then saw software defined storage solution safety distribute object system and it's highly scalable and it's running on general purpose community hardware this slide so so the architectural safe oh we can know from this we can all that we can all that safe the safe cluster is based upon rados and it provides the fair object and the device to users the safety the guiding the guiding population in cloud in in in cloud in cloud as a storage solution this is due to it's open open distribute feature okay safety has different types of common components the most in the most in the most important two components are OSD and the monitor the OSD is the OSD demands store the data handle handle data replication recovery backfilling rebalance and provides some monitoring mission to safe monitor by checking on the safe OSD demand for heartbeat the monitor the monitor the safe monitor maintain the maps or the cluster including in including the monitor map the OSD map and the placement to group mental map and the crash map okay next we will go into introduction about the scale scale is a software software only solution it's in it's installed on the common community servers it's using existing servers local disk and line to create what you said and you suppose it expose the server the local detector disk to the shadow block storage this is the size of shows the architecture of the scale the scale architecture is very much like a safe it has three different types of demand the first is MDM you can see create equal to the monitor in safe it's config and monitor the scale system cluster scale system you can deploy deployment deploy it to the MDM in cluster mode or single mode the another time another OSD the another demand is the SDS you can equal to the OSD in safe cluster the SDS manager the capacity of a single server and the data science the different there are only one SDS in in one server this is different from the OSD you know that you have one OSD for one disk but you have one SDS for one server and the last demand is the SDC the SDC is used to expose the scale scale volumes as the blocker device when you map to scale volumes then you can get a device named the D.U.V. and S.I.N.I.A.B.C. like this device then you can use it the next so commercial storage is solid file solid file is very famous and it's a it's a collection of node hardware and software the solid file storage cluster is a collection of nodes you must have at least four nodes to one to two wrong solid file cluster of course you can add one add node to existing solid file cluster the the the service when to be interrupted during your adding the node and the node the performance and the capacity of the node will be automatically used by the cluster SkyL has different networks for for different usage there the nodes there are 10 GB as a net user to connect the nodes in the storage cluster and you can manager the cluster using one GB as a net and the client can be connected to the server you then 10 GB a network depends on the SCSI it's easy to scout out you can add node to to the cluster okay next we'll give a comparison all these all these storage system first is deployment of safe safe how so many deployment tools you can use the tool deployment to the safe cluster for example safe deploy safe disk Ansible, Parbit, Chaff and C2 C2 is open-sourced by Intel by Intel last month even although there are so many tools you can deploy a safe cluster but for Naio who have no experience with safe it's hard to deploy deploy a safe cluster he it's this is because they will have so many concepts and configure parameters the new person and if the person don't understand the concepts and the configuration parameters he usually get in trouble when he deploy a safe cluster people or anyways safe safe like user friendly tool your friendly tool to help people to do the deployment and the next is deployment of scale there are two ways to deploy and scale cluster the first one way is manual you can deploy it one step by step or the other way is you can use that is interrelation manager manager first you the recommended way is using the interrelation manager first you install the interrelation manager and then you can log in it log in it then you can log in it using Chrome RIE the welcome window that appear in this window you can perform you can do your deployment first you should upload the package you want to in total and the second you should how upload the topology fair the topology fair is you can think it as a crash map in CFO it's defined as the MDM node as the S node STC node and the networks that will be used you know in safe there are two networks the cluster network and and the public network but in scale there are more different more different networks you can define networks used to the communication between MDM and the network you can define a non-network you to the communication between MDM and STS and you can also define a non-network to communication for for STS and STC also you can define the in safe cluster you can define define which OSD or which OSD located in which pool in the same way in scale you can also define which disk between belong to which pool so you can define SSD pool or HDD pool in a cluster up to you up the topology fair isn't CS3 fair you can iterate with Excel or as a file editor and when you have done it you should upload it to the entire manager after that you can begin your into begin your interaction what you want what you need to what you need to do is only a click click the mouse after it stage if you have any error then you can read read down it again and you correct your correct and error and read read do it again okay the last deployment is solid affair solid affair doesn't you you need a nothing in talk in twin star if you want to use the solid affair they are the solid affair story nodes are deployment as an application with solid affair element OS in target and ready to be configured what you need to do is just the initial the storage in the solid affair node and I did it to the cluster then if there are more than four nodes then the cluster then the solid affair cluster it works okay next we'll go okay this table shows the operation and management management all deep all these storage fair system storage system from they can from this table you can you can get that the majority majority are safe the majority operation and management are safe can only be done using the common line interface but the solid affair and the scale how the web UI and the rest gateway to operation and managerial storage system the other way we as I'm for safe but right now it's up to now it is not right day to for usage features and all the methods first let's compare the features of the storage system from this table you can know that are these are these storage systems how the how the basic features safe only safe can spot fair object and the block skyline solid affair can only spot a block of storage but come out to the the commercial storage safer locks or some advanced features for example or the application come come compression and I OPS can guarantee in a cloud environment the I OPS guarantee is very useful because our tenants show the usually the our tenants show the same storage so it's very useful but safe who cannot spot it right now up to now of course you can do it you can do it you will kill but it's not perfect way scale is best has the has the most features all the methods best features he sorry fair cannot control the max LPS or volume but also you the story fair sorry fair system can control the minimum LPS or volume are these the storage the system can be integrated with ice platform for example open stack and then solar but safe can only support to open stack and then solar skyline solid affair can support to open stack hyper V the one more and then solar okay this table compares the volume message in the storage system are these the storage system can support basic volume method for example creating deleting snap shopping resizing and so on but the other big drawback in safe safe doesn't support the queue LPS sighting that is what I have say in last slide that is a QoS support guarantee we next is integrating with the open stack from this side you can know that safe is the best integrating in for open stack it's the same is integrating with open stack component for example including swift singer glass and the Noah but the skyline the story fair can only integrated with the singer the new open stack project Mar Nila can also integrated with the safe okay I trust we in the past we have to a performance test between safe and the sky I this is our test environment we have for we have for nodes for this test each node have have both the SSD and HDD for our test our test and there's deep the configuration detail you can get from this slide we use the FIO and the VD for our test okay this is LPS and the latency we have the result we have to get the left the left histogram shows the LPS all our test so the L is much better than safe in both the SSD pool and HDD pool in SSD pool the LPS or Skyo is more than six times or safe the safe so it's a huge gap between them but in HDD pool they are equal they are closer to each other or cause the skyo is better than safe and the right histogram is latency or the latency or these two to storage system the latency or Skyo is less than safe this latency refers refer to the latency when the cluster is at when the cluster at the full load so it's more this value is much more much is much bigger than the node when then is much bigger than much bigger than the latency when the storage system in the in a normal load so we go into going comparing between between them we use VD Bench to get the data or the the cluster cluster latency at a different node and this one is not very clearly okay I hold description the latency of Skyo no more than the SSD pool or HDD pool is less than one millisecond but for a safer the latency is always more than two or two millisecond the sky the latency of Skyo is smart is a smart line at they are very close they are the value is very close when Skyo at a different load but for safe the latency is increased when the increase the latency is increased when the load is increased okay the right the left his ground shows the cluster stability under certain load we use VD Bench to to to to to to product 600 IO request to the system then we get actual actual IO request IO request we request record the value then we get the left histogram in this histogram you can see that both the sky both the sky are both these two line are both these two have a fraction but the the the safe line is more more instance and more more instant fraction than Skyo so this means that the safe cluster is not so stability when you have have enough load okay this slide shows what happened when scaling out when scan when scaling out are scaling in the left the left the left the chart we show the the Skyo cluster when when adding are deleting a node when deleting a node the black line the black line shows the delete when we deleting a sds node from from Skyo cluster the the cluster have set have several seconds can process any process cannot process any request so the why the request the IO is zero and but so just several seconds after that so the cluster has the has the ability to process the request and the several means later the cluster the cluster the recovery and the the whole LPS is less than the less than the than less than than the cluster we the more I still as the red line so the cluster adding on adding on sds and you see the same like the deleting a node the cluster has a recovery time in this time the cluster can the LPS of the cluster is the flexion is the flexion and I trust the LPS the LPS of the cluster is the recovery and and the marches then the beginning okay the right chart we show we show what will happen when we when the crash map has changed in in safe cluster the right line is the default we have we have not changed the default can can background configuration parameters when we adding are deleting an OSD the the class the crash map changes and the the cluster began to recovery began to recovery began to load balance the the data at this time at during this time there are no the the LPS of the cluster are either very close to zero the that is me the whole car the safe cluster cannot process the client request but if we set these two configuration parameters the OSD max the backfield at set it as one and set OSD recovery max actual as one then we get to the black line when we when the crash map changes the cluster can still can still process the client request so but for for but for for some every every every person you will see for may know we we need to set a recovery limit if you if we have we have to we have to change the crash map but not everyone know which can configuration we we had to set okay what I mean is that the safe one lacks some some tool to help people help people know the no are all its can configurations okay this is a summary first a safe need a more user friendly deployment and the measurement tool for user to deploy a safer cluster and the second safe likes advantage storage feature such as the QoS guarantee the duplication complete completion especially the QoS guarantee is very useful in the cloud storage solution of course the safety is how some some best some good value safety is the best integration for open stack if you configure a safer as the client stand and know a backend you can put you can use you can depend the copy on right to clone feature and you can put instance boot a lot of instance in I told at once is in in second level okay the first safety acceptable for HDD but not good enough for performance disco this we can get this this conclusion from our test on safe and a sky all safe has a lot of communication parameter but likes all relevant instruction and we relation to all safety so safe is not very very easy to configure its to configure its parameter okay that's it thank you