Always On: Highly Available Data Part 3: How Can You Recover Servers in the Cloud?

To recap…

How long can your business survive without your data? How long can your business survive if your people can’t access your business via the internet?

Can you do disaster recovery to the Cloud, so that if your datacenter is cut off, you can failover to a virtual infrastructure, if only for a short while? Is this possible? Is it affordable? If it is, and you can do it and it works, then…why are you in the business of building and running a datacenter at your headquarters?

How can you recover your servers in the cloud? There are many technologies that are now within reach, including host and virtual machine replication (not storage array replication, $$$).

Here is a run-down of those applications and tools that enable you to achieve a recovery time objective of half a day or less, and a data loss of about the same.

Rapidly Changing Technology

I started writing this series a few months back, after a major earthquake took down systems in DC, Virginia and Maryland. Back then, I was exploring many options that have already changed, evolved or been superseded. This is one of the challenges of this area of technology.

We have also put into place a few hybrid backup solutions that have shown some of the challenges that face this area of IT solutions.

I designed a backup and archive system for a customer that uses a major backup software suite, CommVault, to back up their virtual servers and write them to local disk. This is working very well. They also are using a virtual appliance that presents cloud storage as iSCSI LUNs. This is from TwinStrata, and it is called CloudArray Controller. The customer likes this very much. They are using the CloudArray Controller to present a new drive to their file server that they can drag old files and directories to, a manual archiving process. This works great – for uploading. They just got back to me and said that the download speeds are painfully slow, around 200 kbs. That seems to be a glitch, and it is being worked on by the cloud storage provider (using Atmos as the platform). Hopefully we’ll find a throttle that was turned on by mistake. If not, that could present a major challenge to adoption. Other cloud storage platforms might have much better download speeds, but we eliminated the ISP, the cloud array controller software, and CommVault as the culprit. A Speed Test can verify this. (Update: different cloud storage platforms have different “object sizes” – think of the standard IO or block size of storage platforms and file systems – that aren’t published. TwinStrata has been independently testing some of them and has a little black book they use in helping people set up their appliance for best performance. They measure performance as they change the cache block size until they see a significant spike in performance and note that for future tuning reference. Pretty slick, but old-school engineering. I like that.)

The implications of extracting and recovering IT applications of any size are obvious, if you have to download. This is where recovery in the cloud becomes important, but only if there is a high bandwidth, low latency connection between the storage and the computing environment. This is not a widely available capability. I know that Amazon has such connections in certain zones. So architecting a solution requires an understanding of how the computing and storage solutions are interrelated.

Cloud Storage and Cloud Computing

The relationship of these two technologies is poorly understood, and this has to change. This market is ripe for maturing, and until it does, you need someone who has hacked a path through this jungle already. Backup to the cloud – fine. Archive to the cloud – fine. Replication of virtual servers from “home” to the “cloud” – fine, but can be expensive.

Cloud Services is too broad a term, as well.

I can recommend cloud storage. Use cases are for backup and archiving but also regular file serving. The customer can provision storage on their own in conjunction with CommVault backups (Symantec and others are adding “cloud storage APIs” to their software now, but CommVault was the leader here) or TwinStrata use cases (see below) or can engage an online backup service provider.

Online Backup Providers are a Cloud Storage service, specifically around backing up data from customer premises to an undefined but secure location “in the cloud.” This is a managed service in the true sense, and cloud services in the broad sense: scalable, on demand, pay as you go. It can be elastic (shrink if not needed) depending on the contract. Asigra is a major managed service provider software vendor that many services have built their products around. Some of the cloud backup vendors have their own software and their own data centers and some license something like Asigra and buy cloud storage in volume from some provider such as Amazon or Rackspace.

We have another cloud storage technology partner, Nasuni, which provides a Cloud File System. The use case here is a shared file system for multiple offices or locations, with local performance. Backup is NOT a use case. The cloud storage service provider is pre-selected and costs are built into the solution. This has an annual fee. The beauty of this solution is that you get local performance due to caching on local disk. But even more important, this solution provides that single location for file storage that has no file locking or replication issues (other than updating local cache). I don’t know of any solution that does this as elegantly as Nasuni. They have even gone so far as to load test various cloud storage providers. They don’t want to necessarily reveal their current choice because they may change as the market develops and they find a better partner in terms of profit or performance or both. I can tell you that one of the major players asked them to stop the testing because it was bringing down their storage systems! (Define that, you ask – performance was being substantially degraded, is the response, so that customer latency soared.)

We have another cloud storage technology partner, Twin Strata. They offer a virtual or physical appliance that connects to any number of cloud storage providers (not preselected) and presents an iSCSI LUN (block level storage) to the customer servers. TwinStrata has proven to be stable, deliver high performance, and has great support. Use cases for TwinStrata’s technology is broader than just file systems, and may turn out to be one of the most significant cloud storage approaches. The cost of the appliance is by subscription or purchase, and the cloud storage needs to be provisioned separately. Provisioning of cloud storage is the hang-up for the resellers out there – the reseller model for cloud storage is very immature at the moment. Most cloud storage providers want us (the VAR) to buy storage capacity and then resell it. We want to sell it without buying it ourselves. Amazon has a good reseller model that they just rolled out. I hope others will do so soon.

Cloud Computing is another matter, and we have Terremark as a Cloud Service Provider (CSP) partner. They do not have a storage-only offering. Amazon, Rackspace, Nirvanix, iLand – all have computing services offerings of differing natures, but the fundamental nature of this service is essentially Managed Hosting. Instead of putting in a ticket to AT&T, for example, where a team racks and configures a server, its storage, and networking, now a couple of guys sit at a console and do the same thing with a few keystrokes. There is a ton of innovation and new technology behind this, but the bottom line for the CSPs is better margin since the labor and time to provision is reduced, and this should or could be passed along to the customers. The customers also can self-provision, but it is just like an online catalog – fill up your cart with CPU cycles and disk space and pay at the cashier. SLAs and performance expectations must be set. Until then, it is unlikely, as noted in another post, that enterprises will put the mission critical stuff in a “cloud” service, nor will Unix platforms ever be virtualized and managed in the same way.

Conclusion

I intend to focus future postings on point solutions and use case scenarios for storage in the Cloud. I hope that this three part introduction to Highly Available data has prompted some thought and given a background to this emerging technology space.