Flash storage continues to rapidly disrupt the storage marketplace
The desire for more scalable systems is driving changes in the architecture of enterprise storage
To people immersed in storage these trends are well known and have been covered increasingly over the last couple of years (flash has been in mainstream arrays in some form since 2008). I’ve written this article to consolidate my own thoughts rather than as the traditional end of year predictions – ‘if you can’t explain it you don’t understand it‘ is something I believe and one of the principal reasons I enjoy blogging. Despite my interest in storage these events made me realise I’ve taken my eye off the ball and I’m now playing catchup!
Most of the vendor and panel sessions at the conference concentrated on the flash aspects – tiered vs cache, should you use a hybrid or all flash array (or no array at all in Nutanix’s case) although there was also discussion of the scalability of various architectures and technologies. Flash’s key advantage is performance – when compared to spinning disk it’s orders of magnitude faster and much of the current innovation is in trying to overcome its other constraints of cost, lifecycle, form factor etc. It’s not just performance that’s driving the current industry changes however – the desire for greater scalability is driving storage from a centralised model to a more distributed architecture (as described very eloquently by Chris Evans).
Combined these factors imply a major shakeup in the storage industry – it’s going to be a fun few years!
A quick post to say that Netapp have released v2.1 of their Windows MMC management tool, OnCommand System Manager (the download link is the bottom right, NOW account required). This new update brings the usual incremental fixes along with support for Flash Pools, Infinite Volumes (a feature of ONTAP 8.1.1 in cluster mode), and multidisk carrier shelves. It’s also moved to a 64 bit architecture – my ‘upgrade’ simply uninstalled the 32bit version and installed the 64 bit one.
For compatibility the release notes state;
Data ONTAP 7.3.x (starting from 7.3.7)
Data ONTAP 8.0 or later in the 8.0 release family operating in 7-Mode
Data ONTAP 8.1 or later in the 8.1 release family operating in 7-Mode
Data ONTAP 8.1 or later in the 8.1 release family operating in Cluster-Mode
However checking the Netapp compatibility matrix shows that this release is ‘officially’ supported on a smaller number of ONTAP releases, notably ONTAP 7.3.7 or newer (excluding 7.3.4 etc) and 8.03 or newer (excluding 8.01, 8.02 etc). I suspected this was simply timing and that once the new release has been around for longer it would be validated against more ONTAP releases. However I tried it against a few of my filers running a mixture of 8.01p2 and 8.02P6 and found one issue straightaway. The new network checker wouldn’t run against the 8.01p2 controllers as apparently they don’t support the necessary API calls.
If you’re running some of these older ONTAP releases proceed with caution!
It’s that time of year when I book the next London VMUG session into my calendar and rather than my usual ‘here’s the agenda, you should go‘ blogpost I thought I’d recap what the last year has delivered. If this doesn’t convince you that there’s value in attending a free event where you could have learnt all the topics listed below as well as networking with your peers then nothing will. 🙂
If there’s a topic you’d like covered or if you’d like to present something yourself get in touch with the organising commmittee. I’m planning to present at one of next year’s VMUG sessions (it’s about time!) because it’s a user group and real world experience can be gold dust for others to learn from. I’m told we’re a friendly audience!
A thought provoking session on DevOps and service management from Alex Smith. This topic was probably out of many people’s comfort zones but it’s good to learn something new and challenge your thinking.
I could mention the giveaways (iPad, Fusion-IO card, t-shirts, AppleTV etc) and the free beers afterwards, the fact we had at least five VCDX’s presenting and the live labs from EMC, VMTurbo, and Embotics etc but you’re already sold right?
In this article I’m going to talk about Zerto, a data protection company specialising in virtualized and cloud infrastructures who I recently saw as part of Storage Field Day #2. They’ve presented twice before at Tech Field Days (as part of their launch in June 2011 and Feb 2012) so I was interested to see what new developments (if any) were in store for us. In their own words;
Zerto provides large enterprises with data replication solutions designed specifically for virtualized infrastructure and the cloud. Zerto Virtual Replication is the industry’s first hypervisor-based replication solution for tier-one applications, replacing traditional array-based BC/DR solutions that were not built to deal with the virtual paradigm.
When I first heard the above description I couldn’t help but think of VMware’s SRM product which has been available since June 2008. Zerto’s carefully worded statement is correct in that SRM relies on storage array replication for maximum functionality but I still think it’s slightly disingeneous. To be fair VMware are equally disingeneous when they claim “the only truly hypervisor level replication engine available today” for their vSphere Replication technology – marketing will be marketing! 🙂 Later in this article I’ll clarify the differences between these products but let’s start by looking at what Zerto offers.
Zerto offer a product called Zerto Virtual Replication which integrates with vCenter to replicate your VMware VMs to one or more sites in a simple and easy to use manner. Since July 30th 2012 when v2.0 was released it supports replication to various clouds along with advanced features such as multisite replication and vCloud Director compatibility. Zerto are on an aggressive release schedule given that the initial release (which won ‘Best of Show’ at VMworld 2011) was only a year earlier but in a fast moving market that’s a good thing. For an entertaining 90 second introduction which explains what if offers better than I could check out the video below from the companies website;
Just as server virtualization opened up possibilities by abstracting the guest OS from the underlying hardware so data replication can benefit from moving ‘up the stack’ away from the storage array hardware and into the hypervisor. The extra layer of abstraction lifts certain constraints related to the storage layer;
Array agnostic – you can replicate between dissimilar storage arrays (for example Netapp at one end and EMC at the other). For both cloud and DR scenarios this could be a ‘make or break’ distinction compared to traditional array replication which requires similar systems at both ends. In fact you can replicate to local storage if you want – if you’re one of the growing believers in the NoSAN movement that could be useful…
Storage layout agnostic – because you choose which VMs to replicate rather than which volume/LUN on the array you’re less constrained when designing or maintaining your storage layout. When replicating you can also change between thin and thick provisioning, or from SAN to NAS, or from one datastore layout to another. A typical use case might be to replicate from thick at the source to thin provisioning at the DR location for example. There is a definite trend towards VM-aware storage and ditching LUN constraints – you see it with VMware’s vVols, storage arrays like Tintri and storage hypervisors like Virsto so having the same liberating concept for DR makes a lot of sense.
Zerto goes further than just being ‘storage agnostic’ as it allows further flexibility;
Replicate VMs from vCD to vSphere (or vice versa). vCD to vCD is also supported. This is impressive stuff as it understands the Organization Networks, vApp containers etc and creates whatever’s needed to replicate the VMs.
vSphere version agnostic – for example use vSphere 4.1 at one end and vSphere 5.0 at the other. For large companies which can typically lag behind this could be the prime reason to adopt Zerto.
With any replication technology bandwidth and latency are concerns as is WAN utilisation. Zerto uses virtual appliances on the source and destination hosts (combined with some VMware API calls, not a driver as this article states) and therefore isn’t dependent on changed block tracking (CBT), is storage protocol agnostic (ie you can use FC, iSCSI or NFS for your datastores) and offers compression and optimisation to boot. Zerto provide a profiling tool to ‘benchmark’ the rate of change per VM before you enable replication, thus alllowing you to predict your replication bandwidth requirements. Storage I/O control (SIOC) is not supported today although Zerto are implementing their own functionality to allow you to limit replication bandwidth. Today it’s done on a ‘per site’ basis although there’s no scheduling facility so you can’t set different limits during the day or at weekends.
VMware’s vSphere is the only hypervisor supported today although we were told the roadmap includes others (but no date was given). With Hyper-V v3 getting a good reception I’d expect to see support for it sooner than later and that could open up some interesting options.
Zerto’s Virtual Replication vs VMware’s SRM
Let’s revisit that claim that Zerto is the “industry’s first hypervisor-based replication solution for tier-one applications“. With the advent of vSphere 5.1 VMware now have two solutions which could be compared to Zerto – vSphere Replication and SRM. The former is bundled free with vSphere but is not comparable – it’s quite limited (no orchestration, testing, reporting or enterprise-class DR functions) and only really intended for data protection not full DR. SRM on the other hand is very much competition for Zerto although for comparable functionality you require array level replication.
When I mentioned SRM to the Zerto guys they were quick to say it’s an apples-to-oranges comparison which to a point is true – with Zerto you specify individual or groups of VMs to replicate whereas with SRM you’re still stuck specifying volumes or LUNs at array level. Both products have their respective strengths but there’s a large overlap in functionality and many people will want to compare them. SRM is very well known and has the advantage of VMware’s backing and promotion – having a single ‘throat to choke’ is an attractive proposition for many. I’m not going to list the differences because others have already done all the hard work;
Looking through the comparisons with SRM there are quite a few areas where Zerto has an advantage although to put it in context check out the pricing comparison at the end of this article; NOTE: Since the above comparison was written SRM v5.1 has added support for vSphere Essentials Plus but everything else remains accurate
RTO in the low seconds rather than 15 mins
Compression of replication traffic
No resync required after host failures
Cloning of the DR VMs for testing
Point in time recovery (up to a max of 5 days)
The ability to flag a VMDK as a pagefile disk. In this instance it will be replicated once (and then stopped) so that during recovery a disk is mounted but no replication bandwidth is required. SRM can’t do this and it’s very annoying!
vApps supported (and automatically updated when the vApp changes)
vCloud Director compatibility
If you already have storage array replication then you’ll probably want to evaluate Zerto and SRM. If you don’t have (or want the cost of) array replication or want the flexibility of specifying everthing in the hypervisor then Zerto is likely to be the best solution.
DR to the Cloud (DRaaS)
Of particular interest to some customers and a huge win for Zerto is the ability to recover to the cloud. Building on the flexibility to replicate to any storage array and to abstract the underlying storage layout allows you to replicate to any provider who’s signed up to Zerto’s solution. Multisite and multitenancy functionality was introduced in v2.0 and today there are over 30 cloud providers signed up including some of the big guys like Terremark, Colt, and Bluelock. Zerto have tackled the challenges of a single appliance (providers obviously wouldn’t want to run one per customer) providing secure multi-tenant replication with resource management included.
Often this is what it comes down to – you can have the best solution in the market but if you’re charging the most then that’s what people expect. Zerto are targeting the enterprise so maybe it shouldn’t be a surprise that they’re also priced at the top end of the market. The table below shows pricing for SRM (both Standard and Enterprise edition) and Zerto;
Zerto Virtual Replication
$195 per VM
$495 per VM
$745 per VM
As you can see Zerto costs a significant premium over SRM. When making that comparison you may need to factor in the cost of storage array replication as SRM using vSphere Replication is severely limited. These are all list prices so get your negotiating hat on! We were told that Zerto were seeing good adoption from all sizes of customer from 15VMs through to service providers.
I’ve not used SRM in production since the early v1.0 days and I’ve not used Zerto in production either so my thoughts are based purely on what I’ve read and been shown. I was very impressed with Zerto’s solution which certainly looks very polished and obviously trumps SRM in a few areas – hence why I took the time to investigate and write up my findings in this blogpost. From a simple and quick appliance based installation (which was shown in a live demo to us) through to the GUI and even the pricing model Zerto’s aim is to keep things simple and it looks as if they’ve succeeded (despite quite a bit of complexity under the hood). If you’re in the market for a DR solution take time to review the comparison with SRM above and see which fits your requirements and budget. Given how comprehensive the feature set is I wouldn’t be surprised to see this come out on top over SRM for many customers despite VMware’s backing for SRM and the cost differential.
Multi-hypervisor management could be a ‘killer feature’ for Zerto. It would distinguish the product for the forseeable future (I’d be surprised to see this in VMware’s roadmap anytime soon despite their more hypervisor friendly stance) and needs to happen before VMware bake comparable functionality into the SRM product. Looking at they way VMware are increasingly bundling software to leverage the base vSphere product there’s a risk that SRM features work their way down the stack and into lower priced SKU’s – good for customers but a challenge for Zerto. There’s definitely intriguing possibilities though – how about replicating from VMware to Hyper-V for example? As the use of cloud infrastructure increases the ability to run across heteregenous infrastructures will become key and Zerto have a good start in this space with their DRaaS offering. If you don’t want to wait and you’re interested in multi-hypervisor management (and conversion) today check out Hotlink (thanks to my fellow SFD#2 delegates for that tip).
I see a slight challenge in Zerto targeting the enterprise specifically. Typically these larger companies will already have storage array replication and are more likely to have a mixture of virtual and physical and therefore will still need array functionality for physical applications. This erodes the value proposition for Zerto. Furthermore if you have separate storage and virtualisation teams then moving replication away from the storage array could break accepted processes not to mention put noses out of joint! Replication at the storage array is a well accepted and mature technology whereas virtualisation solutions still have to prove themselves in some quarters. In contrast VMware’s SRM may be seen to offer the best of both worlds by offering the choice of both hypervisor and/or array replication – albeit with a significantly less powerful replication engine (if using vSphere Replication) and with the aforementioned constaints around replicating LUNs rather than VMs. Zerto also have the usual challenges around convincing enterprises that as a ‘startup’ they’re able to provide the expected level of support – for an eloquent answer to this read ‘Small is beautiful’ by Sudheesh Nair on the Nutanix blog (who face the same challenges).
Disclosure: the Storage Field Day #2 event is sponsored by the companies we visit, including flight and hotel, but we are in no way obligated to write (either positively or negatively) about the sponsors.
While upgrading my home lab recently I found myself reconsidering the scale up vs scale out argument. There are countless articles about building your own home lab and whitebox hardware but is there a good alternative to the accepted ‘two whiteboxes and a NAS’ scenario that’s so common for entry level labs? I’m studying for the VCAP5-DCD so while the ‘up vs out’ discussion is a well trodden path there’s value (for me at least) in covering it again.
There are two main issues with many lab (and production) environments, mine included;
Memory is a bottleneck and doubly so in labs using low end hardware – the vCentre appliance defaults to 8GB, as does vShield Manager so anyone wanting to play with vCloud (for example) needs a lot of RAM.
Affordable yet performant shared storage is also a challenge – I’ve used both consumer NAS (from 2 to 5 bays) and ZFS based appliances but I’m still searching for more performance.
When I built my newest home lab server, the vHydra I used a dual socket motherboard to maximise the possible RAM (up to 256GB RAM) and used local SSDs to supplement my shared storage. This has allowed me to solve the two issues above – I have a single server which can host a larger number of VMs with minimal reliance on my shared storage. The concepts are the same as solutions like Fusion-IO aim to do in production environments but mine isn’t particularly scalable. In fact it doesn’t really scale at all – I’ll have to revert to centralised storage if I buy more servers. Nor does it have any resilience – the ESXi server itself isn’t clustered and the storage is a single point of failure as there’s no RAID. It is cheap however, and for lab testing I can live with those compromises. None of this is vaguely new of course – Simon Gallagher’s vTardis has been using these same concepts to provide excellent lab solutions for years. Is this really a poor man’s Fusion-IO? There’s nothing like the peformance and nothing like the budget but the objectives are the same but to be honest it’s probably a slightly trolling blog title. I won’t do it again. Promise! 🙂
If you’re thinking of building a home lab from scratch consider buying a single large server with local SSD storage instead of multiple smaller servers with shared storage. You can always scale out later or wait for Ceph or HDFS to elimate the need for centralised storage at all…
Tip: It’s worth bearing in mind the 32GB limit on the free version of ESXi – unless you’re a vExpert or they reinstate the VMTN subscription you’ll be stuck with 60 day eval editions if you go above 32GB (or buying a licence!).