• dynamo db ssd screen-grab

Interestingly the most visible SSD proponent and database service is none other than Amazons AWS service DynamoDB. Launched earlier this year, a database service which can be scaled at the turn of the knob, whilst offering performance, availability and the NoSQL feature-set. Considering the whole premature klonking-out with small-block, highly-randomised IO that hits the SSD controllers, you’d think a very ambitious undertaking.

SSD’s are great, but not a panacea for all workloads, in their raw state; especially for write-skewed database workloads. One of the largest causes of headaches for SSD Controller manufacturers and those looking to optimize workloads on SSD’s is Write Amplification, otherwise known as the variable amount of operations or re-writes of the original size that have to occur before a write is actually processed / stored by the disk.

The conversation reminded me of a post by Todd Hoff who attended an AWS session of “DynamoDB for Developers”, and wrote an interesting line:

Initially they ran into performance problems due to SSD garbage collections cycles

This is inevitable. In my VMUG talk from last week, I refer to the Controller, and rather its FTL (Flash Translation Layer) as the black box of the device, where all the secret sauce of the unit lives. This isn’t by mistake, as all the IP around the type of algorithms to use for wear levelling, garbage collection or any internal process resides within.

• SSD controller

A single SSD within a single host, can cope with a lot of write-abuse, as given even a few minutes of ‘idle’ time, can effectively “defragment” its block + page layouts, and thus provide a buffer for the next onslaught. Although en-masse, where AWS operates, this simply will not work. As you will have many thousands of users and transactions in the hundreds of millions (+speculation), the drive will not have time to rest and carry out the vital to its survival, levelling and Garbage Collection.

As the folk over at Tintri have shown, when SSD’s are pummelled to a point of saturation OR effectively outside the boundary of when the internal algorithms perform cleanup, the performance or rather the predictability of said performance is significantly diminished. Below you can see latencies of nearly 300ms for an SSD device.

• source: tintri blog

When you’re operating a on a significant scale, you can’t have “unknown” hold you back, so that’s why (+speculative mode) Amazon worked directly with the manufactures of their SSD suppliers to have a say in:

• When and how GC kicks in
• RAID design across individual cells (similar to vRAID within Violin)
• Granularity of block mapping
  • How blocks are read+cached+written, and aligned DIRECTLY for DynamoDB’s block-sized, in order to not necessarily alleviate, but minimise re-writes
• Locality of writes

The last being most interesting, and I believe an absolute must for the future of SSD to have an enduring life in the enterprise. Applications, or rather file-systems and thus API’s should allow for the ability to direct or at least specify the inter-relationship of data being written. Simply dumping blocks and forgetting about them is simple, but has dire consequences for the life of the storage medium in this case. Surrounding writes, or data operations with meta about them allows for solutions such as DynamoDB to exist and flourish. This requires a very tight integration between the software and hardware. There’s a reason Apple is one of the largest NAND / SSD suppliers (direct to consumer) out there, and why they bought Anobit, the Flash-controller firm last year. Go ponder.

Last night I read a post by Aditya over at Savvis that looks at “horizontal application scaling into the cloud”, and what such a solution will encompass.

I take an issue not with the article, but the concept itself in the current incarnation when its applied to the area of enterprise applications.

In order for a workload to be able to ‘burst’, and as helped by the dictionary it must be able to “break open or apart suddenly and violently”. When was the last time you’ve seen the tightly-coupled layers of any application be able to be broken apart without induced downtime? There are interdependencies not only surrounding applications but the infrastructure supporting them.

The utopia that seems to dominate the cool-aid drinking cloud echo-chamber is an all-public consumption model, where public-only / single-provider is the method for all IT service delivery. Reality is, in privacy, process and control-centric government and enterprise environments some data will always remain in a highly controlled and self-managed environments.

For me, CB’s most realistic use-case is going from logical-local or private premises to an external provider, what Forrester terms ‘virtual-private’.

In order for CB to be possible, I see two camps developing:

1. The interface between infrastructure, management and monitoring must be very clearly defined to allow for tight integration between all components and thus provide a higher awareness of actions being carried out. Such as performance re-balancing, or disaster mitigation process instantiation

• This is where over the last 2 years we’ve seen companies playing the converged infrastructure card, and spend copious $’s on R&D, integration, testing and product delivery

2. Applications must be designed with an “expect-to fail” approach – a discussion which if memory serves right was between @reillyusa and @mreferre. In summary, by taking care of failure at a higher layer, specific to the knowledge of your application, it’s cheaper than having to design an extensive-feature-rich, all use-case covering container

Both have their merits in todays market, and for shops which are not tech-savvy and have extensive legacy baggage or who’s main purpose isn’t IT differentiation, the first choice is a step in the right direction. On the other hand, if you are developing applications for the new world, creating scalable services and are full of zeal to produce the best possible architecture in your niche, then the latter choice is IMO a good approach.

Whatever the correct choice for your bussiness, one constant remains, you need to begin at the application.

Not saying it’s impossible, but uploading, converting, VMDK/ OVF/AMIs, starting them, adding IP’s to loadbalancers… brr – is a joke at best.

— Roman Tarnavski (@romant) April 17, 2012

• Endurance vs Performance vs FTL Capability

The paper, written by Laura M. Grupp, John D. Davis and Steven Swanson does make some very valid points, and more importantly it is backed with empirical data – which I adore.

in summary

The latency will be affected by the diminishing size of the manufactured die, and as we approach 12nm, the performance as measured in read response time will increase from being ~20µs to ~25µs, or have a delta of 0.005ms.

Most significant write throughput decrease has already occurred between the NAND addressable capacity increasing to 256GB. When you start looking at 1TB to 4TB transition, we’re losing less than 250MB/s. Easily alleviated with multiple smaller cards.

Most evident sentence which should be used to ground the academic nature of the article is:

fastest hard drives can sustain no more than 200 IOPs, and the slowest SSD configuration we consider achieves over 32,000 IOPs

So if I were to lose say 15/20 or even 30% of that performance in IOPS per addressable card, and the response time was a hideous 30µs even … am not sure that’s quite such a bad trade-off for a 4TB of NAND love.

clusters to the rescue

The gloom was very well painted at the beginning of the paper, but as you get beyond prophetic statements and consider the generational shifts that have appeared between tape, myriad of rotating disk, and now Flash for primary storage; you start to realise a single constant. Flash technology is not here to replace spinning disk, but to augmented the delivery of application IO.

As an application stands today, it still hasn’t changed its consumption model of the underlying block IO, so for the time-being, fundamental principles of smaller, purpose-chosen storage-devices working as part of a larger RAID-set will continue to result in a significant performance uplift. There are a number of Flash drive and array manufacturers today already designing clusters at not only intra-device at the chip-level, but also across multiple cards in the OS. Each with its own benefits, and each surpassing the metrics Laura et. al claim will slow the adoption of Flash.

For those that read The Innovator’s Dilemma may remember that “performance oversupply is an important factor driving the transition” – which we’re flooded with on a monthly basis with IOPS records. Only difference now, with excess idle computational capability within the arrays, coupled with smarts around Hierarchical Storage Management (HSM), we can deliver that IO faster than ever before with Flash, but also tap the blended sets of primary storage based on old variations of spinning disk. All provided to the application as a single, seamless stream of superfast bytes.

Lets not split hairs, Flash is here to stay, and in its current incarnation will continue to evolve, reduce latency to microseconds, and drive throughput beyond what a practical amount of duct-taped-together spinning disk can amass.

References

1. The Bleak Future of NAND Flash Memory – link

When we start looking at availability of mechanical drives or most components inside a server, we first refer to the MTBF.

The consumer-grade disks are generally rated at several thousand hours, whereas enterprise-grade drives typically see 1.3million+ hours of operation.

Due to the differences between spinning and SSD disk, it isn’t always appropriate to use MTBF as a measurement of the drives availability.

The NAND memory cell within the SSD doesn’t act in the same fashion as a spinning-disk block-abstraction. It must first be erased, prior to a write being carried out; unlike the spinning medium where a simple overwrite is possible.

The bits within a NAND cell are registered by electrons moving through a membrane/tunnel-oxide. It is responsible for keeping those electrons in check.

• Program / Erase Cycle - click to enlarge

Gradually, over many program/erase operations the oxide starts to wear out. It is this event that should be avoided, as it will slowly start to cause single-bit errors, and cell locations that expect a 0 will be met with a friendly 1.

For this reason, the Flash Transformation Layer, responsible for all the internals of NAND management & carries out wear-leveling operations which evenly spread the load across hopefully all NAND cells.

Whilst looking at SSDs, there are SLC, MLC, and recently eMLC based flash memory. Difference being the amount of bits that are stored in the individual NAND cells. SLC has a single, whilst MLC/eMLC store two bits. In a lot of cases they’re actually exactly the same physical die, and what separates SLC from MLC/eMLC is the aforementioned FTL, amount of reserved space for the load balancing, and capability of the onboard ECC. This results in tunnel oxide in an SLC-rated drive having a typical life of around 100,000 program/erase cycles and 10,000 for MLC.

Now that you can see, that -most- of the wear on the drive occurs during cells actually being written to, and not idle time, given lack of moving components. With nothing changing, the larger the drive coupled with better FLT, the longer the drive will endure.

The reality check comes in the form of manufacturing processes, and as AnandTech points out, with the reduction in die size, results in extra sensitivity required to read each bit – there’s an inevitable hit in individual endurance of the cells. Yet the performance has been clearly increasing with each generation of die-size reduction, which is a result of the gain achieved in the controller power and thus any ECC coupled with greater reserved capacity on the disk has a net effect of – higher performance, and total better longevity.

• Endurance vs Performance vs FLT Capability

NAND Endurance

Instead of looking at a mechanical, time-driven metric, the best way to calculate endurance of the drive is stress the oxide tunnel, and see how many times it can be trespassed, which translates to the amount of bits that can be written.

A standard that only several manufacturers have started to adopt is the JESD218 which gives an Endurance Rating in Terabytes Written (TBW). It also nicely classes client vs enterprise workloads where something as simple as the Power on state is measured (among others… see references).

Client

• 40ºC
• 8hrs/day

Enterprise

• 55ºC
• 24hrs/day

based on a use scenario in which the SSDs are actively used for some period of time during which the SSDs are written to their endurance ratings

Most importantly the standard is not using just time as a function by which to measure, but the actual endurance level. Therefore when you see a 20PB “TBW Enterprise” rating on a drive, you can quickly determine that the endurance of the drive over a 5 year period offers an ability to write 11 Terabytes per day [20*1024/(5*365)].

Next time you’re picking a drive, be weary of the correct metric being used to show you how long the unit will actually last. Knowing that each application is different, with its own workload profile – you will be able to easily figure out how long the said drive will last.

References

1. SSD is Not the Best Way To Use Flash Memory in Storage – link
2. Micron’s ClearNAND: 25nm + ECC, Combats Increasing Error Rates – link
3. Solid-State Drive (SSD) Requirements and Endurance Test Method – link
4. SSDs… heaven or hell? The pros and cons of using NAND flash memory – link

• Happy Germans | Nora + Christoph

As my eyes were pried open, I uttered a somewhat able ‘hola’ to the newly found adjacent bunk-bed companions. To which one of them nicely reciprocated – “You look damaged”. Now that’s how you say “hello”!

All started with having nearly missed the flight in Buenos Aires, it was a pleasure to embark on a long-ish flight to Lima. The service was superb, food enticing, and surprising of all – metal cutlery. It also helps when the flight crew is pleasant to look at. Arrival at the hostel was post the mandatory airport-taxi haggling. Another pleasant surprise awaited; the Kokopelli hostel is situated in the Miraflores district, which for the un-initiated is one of the largest tourist traps and entertainment areas within Lima or as I refer to as cleaner version of “Sydney’s ‘Cross”. The whole place was alive with music and energy as a result of a few thousand back-packers; with fitting entertainment within an area not larger than a square kilometre.
The first stop was a pizza-sponsored dinner. Upon walking through the little street of restaurants and clubs, the abundant maître d’s from each establishment do their best to hustle you inside. Having had the pleasure of enjoying a so-called ‘pizza’ in Buenos Aires, I was hesitant to subject myself to it again, to be fair, if you want 3cm of nothing but a cheese with an oil volcano on the plate, then Argentina has the best pizza in the world. Fortunately Peru didn’t miss the memo on the preparation, and appropriate calorie level of this fine culinary specimen, and delivered on a somewhat passable pizza.
Following destination was a salsa club, about 5 doors down. Here I witnessed feats of impressive flexibility and gracefulness of middle-aged men dancing with such zeal, that it was simply a pleasure to watch. Some time has passed, during which further blood dilution occurred, which had a direct correlation with my self-confidence, and supposed ability to perform same moves. The locals were very helpful in instructing a gringo in the art of the Salsa for the duration of the stay. The bed-crash at the hostel occurred at an estimated 5am.

–

Meet Christoph, the transport economist and Nora, the dental explorer. Once Chris explained to me what he does, whereby he deals with a branch of economics with a mixture of civil and economy-incentive engineering, I simply wished our own CityRail employed these ‘German Efficiency Engineers’. Chris, how about you come to work in Oz? Our government needs you.
Nora on the other hand, spent the morning picking the most minute flaws in my teeth. It was a great way to incentivise me not to open my mouth for the rest of the day.
Once the ritual of breakfast was concluded, we all banded together to have lunch; so we headed into the city. Being easily impressed, I was delighted that the bus into town, didn’t have to share the same road as the rest of the horn-loving, lane-ignorant road citizens of Peru. The bus got its own, dedicated lane, separated by median strips; a simple way of guaranteeing QoS. For lunch, we had the famous raw-fish Ceviche, some ill-hydrated urine looking warm beverage that was quite sweet, and surprisingly tasty. Whilst everyone went for the safe bet, chicken + rice, I ended up having some what I later discovered not overly well cooked local octopus.

• Ceviche – marinated fish. Tasty!

Now that we had some sustenance to go on, we proceeded to walk about town. Having decided to spend a small fortune of Peru Sol’s on some tramp stamps also known as material country flags for being vagrants in other lands, we then had a democratic discussion as to the location of each flag upon our bags, especially the USA one, perhaps I can stitch it out of sight, to the bottom.
We witnessed a holy day celebration of Our Lady of Mercy outside the local church, and officially marked it a success and concluded the day with a 1 Sol wine tasting session amidst some underwear displays.

• cheap wine consumption

• 3 amigos discussing Patagonia over some Calafate Sours

Especially if you consider that the main walk[s] consist of half a dozen variations to a set track, the paths themselves in most places are no more than 1m in width. Pushing through a hundred smelly backpackers as you traverse the peaks of this beautiful terrain would be a horrible horrible thing … that’s why the best time to visit is pre-season! (shh. . don’t tell anyone)

There exist many pros on which I’ll concentrate first – in exploring the Torres Del Paine Mountain range pre-season. First is the lack of Germans with their walking poles behind every bush. History has shown, its never good to have them band together and march great distances through a foreign land.

The next really depends on how luxuriously you wish your stay to be. There’s a number of hotel$ (privately owned), camping sites at literally every trek intersection, as well as a small number of refugios. Which are nothing more than … well, a bed, roof, kitchen, + fireplace to air out your socks, these are also privately owned although the stay is malleable starting at $30odd USD per person. Simple benefit for the pre-seasoned traveller is space. In a typical high-season, bookings seem to be essential due to the firehose of willing participants. We ended up staying in a bunk-bedded room with a Korean gentleman who I can only assume was full of Godly praise for his 2-minute-noodles as he devoured them slurpingly and Roman-awakedly.

First night we stayed at the campsite alongside Hosteria Las Torres, as they wanted to charge over $300USD for the night, per person. Not to be put off, the campsite is well trimmed and looked after, so spending the night away beneath the Torres was actually very comfortable. By venturing into the Las Hosteria for a sip of the local brew, we met Ignacio, the “Aussie Lover”. Formerly a stock broker from Santiago, followed by oyster shucker in Broom, and resident of Bondi; is now a guide within Torres Del Paine.

The first days trek of the W-course is around 11km by foot, we elected for a different approach. Horses! Throughout the ‘off-season’ the Gauchos (people who beat these beautiful animals into submission, aka cowboys) let the horses roam free throughout the park. They started collecting them only days prior to our arrival. It would be remiss of me to not refer to them as ‘rock climbing horses’ or as Rob put it “mountain ponies”. The feats these wonderful beasts performed were incredible. Throughout the ‘horses-only’ trek, which is closer to the edge of the river, and in my opinion a significantly better view than achieved by plebs on foot; our companions were able to scale near perfectly flat rock faces of nearly 30º (I had a protractor with me…) with surprising ease. Have never seen a horse do anything remotely close, mind you, this was officially my second ever horse ride.

• rockclimbing pony

The vistas available within this part of the route (Las Torres -> Los Cuernos) spoilt us considerably as for the hours after dismounting, the rain and wind conspired together, and decided to test our gear; everything held up extremely well, apart from my atrociously designed Salomon Cosmic 4D GTX. Bad bad shoe. They’re very pretty, just don’t buy them, at least not if you want to actually use them.  Having finished the walk, the views available on this leg were still the best part of the whole week. Make sure you spend considerable time, and camera memory here. Unfortunately I couldn’t hand hold my camera, whilst attempting to detract my four-legged “sure-foot” (name credit to Rob) from sniffing the butt of the horse ahead, and not smacking me into the lower branches of trees. Photography-wise, the plebs win here.

By the time we arrived at Campento Italiano, I could have cooked several packets of spaghetti from the water in my shoes. Mainly this is a campsite, with a single wooden shack with a dirt floor, to shelter the pilgrims on their way through the park. We found a very friendly, fast talking, brightly coloured (I thought my jacket was bad!) French-man with A.D.D., we proceeded to borrow a bunsen burner he had in hope of heating up some canned ravioli, to somewhat offset the cold gust of wind and now splattering of rain.

Upon drying my socks on the rocks next to a fire, and clamouring into the tent before the next wave of ice-cold water deluge from the trees hit me, I rested.

• please send help … soon

The stormy night, was followed by a more impressive morning. Donning the daypack filled with camera gear, we rushed up to Mirador Frances, to see remnants of a glacier from the lookout, although after witnessing the mighty Perito Morino the week prior, it quickly underwhelmed. Am not sure as to why it bears ‘French’ in the name of the valley; there is not as much dirt, didn’t see any dog faeces or foreigner-disapproving, baguette-wielding locals pretending not to speak English as you ask them for help at 2am… but I digress. Even the French-man from the night before was unimpressed.

It is said that during the bloom of the season, the French Valley is one of the most beautiful sights due to its microcosm of flora from around the park.

Continuing along the path, this time towards Refugio y Compamento Pehoé, we were once again greeted by some of the friendly wind. Upon arrival, only slightly wet this time, we found solace in seeing walls, and an actual floor, a roof and to further top it off – a fireplace! It appears about 10 more people were sold on the same idea at $35USD, so an evening of pasta, tuna, chocolate, and sharing of travel stories with the fellow wet-feeters ensued.

• windy day

The W-trek can be started at either side, right->left, or in reverse. At the end of our W, was the administration building; highly discouraged by other travellers due to being a trip of boredom, although quite an easy 18km walk.

As we left the protection of the refugio, en route to the administration, we spent over an hour within about 100 metres of the refugio itself; the weather finally turned, even if for only two hours, to let us capture some of the beautiful landscapes that exist within the park. Most notably right near the refugio; the owners sure know how to pick a good spot!

• outside the refugio

Arriving at Administracion CONAF, the wind almost pushing us out from the park. Continued to belt at full strength. Lack of open-season, resulted in this case the local hotel/hostel to be shut, with the only option 2mm of nylon + polyester tent to protect us at night from the whistles and face-shredding wind.  Had no choice, but to whore ourselves out to passing cars making way in the direction of Puerto Natales. Luckily for other 99,998 gringos that will visit Torres Del Paine this year, not only are the local drivers, but other tourists also happy to give you a lift. Unfortunately hardly anyone is travelling in the same direction, and duration. After just over 40 minutes of excitedly running up to stopping cars, with eyes of a 6 week-old hungry puppy; finally managed to snag a lift – on the back of a ute. Not only was the owner driving his whole family, but he still slowed down, and helped us. I take my many hats off to the driver.

• back of the ute & dust-filled road back to Puerto Natales

Those are some of the more positive things about coming to the park before the season officially opens. What any visitor has to realise is that one of the most unpredictable parts, is the weather. Not only overnight, but within 30 minutes can go from postcard-like scenery to clouds that attempt to ruin the holiday by pelleting strategically placed rain drops into the open crevices within your poorly designed hiking boots. Other point to consider is food. If you visit when none of the refugios nor hotels are open, you will have to carry enough to sustain your ever-eager metabolism over the duration of the trek.

As an aside, the day following our return to Puerto Natales, was the Chilean Independence day. Complete with a parade, BBQ’s at every corner, and festivities for kids within the fire station. It was great to retreat to one of what became the warmest place to lunch when in town, Cafe Toore with its own “Patagonian Consultant”, Sergio, and as you can see in the photo, his daughter.

• our friendly consultant Sergio and his daughter

Consuming meat for breakfast, lunch and dinner at every opportunity, Dieciocho (Chile National / Independence Day) on September 18th was close to my stomach. Devouring a couple anticuchos, followed by some Mote con huesillo, it was an amazing conclusion to the first leg of this South American adventure.

• mmm … yummy local food & drink

"basics" I took

The journey around Torres Del Paine, and adjoining parks is to take nearly two weeks. The vast majority of them will be on foot, thus consideration must be given to the weight of the total pack(s), as there are (unfortunately) going to be no sherpas.

Weather report for this time of year, idled within 2-3ºC below 0 for the night, and a few degrees higher for daytime. Which necessitated the number of thermals throughout the trip, coupled with the self-imposed weight restriction as well as possible lack of showers; more thermals = less smelly experience.

There are many lists online of what people have managed to bring with them along for their trek. To add to the noise, I wanted to provide an OCD technologists view.

Main pack is the Genesis 70, which comes with a very useless daypack, or one you can use to place your hopes and wishes into. Hence you see me bringing the Kata bag instead. Not only is the latter going to be used for housing the camera equipment + laptop, but also re-purposed as a daypack.

Here’s the complete list of the crap thats coming with me. Most is self-explanatory:

Containers

• backpack 1 – MacPac Genesis 70
• backpack 2 – Kata Sensitivity V
• medical kit
• 2-man Tent
• Mattress
• Sleeping bag – Mountain Hardwear Phantom

Clothes

• 1 x Merino – 280 GSM Midlayer
• 2 x Merino – 150 GSM Underlayer
• 3 x Merino Long-johns
• 1 x compression socks (for flights)
• 8 x sets of thin socks
• 8 x undies
• 1 x Goretex wind/water jacket
• 1 x REI Waterproof pants
• 1 x glove inserts (for warmth)
• 1 x gloves, thin but water + wind resistant
• 1 x boardies
• 2 x bonds top
• 1 x jeans
• 1 x t-shirt
• Boots, Salamon Cosmic 4D GTX
• Runners, Ascics
• Sandals, Keen
• 1 x Bose QC 15 (not for the hike itself…)
• Polarised glasses & hard case
• 2 x Pocket Handwash – Sea To Summit
• 1 x Body Wash - Sea To Summit
• Sunblock & Insect Repellent in one - Sawyer
• Multifunction Knife – Leatherman Kick
• Firestarter – exotac nanoSTRIKER
• 20m Parachute cord
• Katadyn MyBottle
• 1 x Micro-fibre Towel
• 25L waterproof bag – used to house all the clothes inside the backpack
• 6 x Gladwrap sandwich bags for waterproofing some electronics
• 1 x QuickDraw Set
• 1 x Whistle (hey… you never know)
• 1 x Washable Electric Shaver – Braun 7 – highly recommend it

Technology

• 2 x HDD 320GB iomega
• 2 x HDD 500GB Western Digital Passport
• 1 x HDD OCZ Vertex 2 SSD, great for the Lightroom catalog, and currently working-on photos
• Workflow:
1. Each evening, the photos from my and Rob’s cameras get copied to one of the marked hard-drives
2. Then copied onto the replica drive
3. Only then are the CF / SD cards cleared within the cameras
4. Any work, sorting, adjusting of photos is then done in Lightroom against one of the drives
5. Upon completion, the drives are once again synchronised
6. The two drives remain completely separate from there on, at no point do they sit in a single backpack, or room (apart from when I’m there)
7. Fortunate to have a base in Buenos Aires to go back to, so the iomega’s get dumped to the WD’s upon return. One of the WD’s gets taken with, other remaining
• No one likes losing photos, especially me
• USB Cables 2.0 + 3.0 (+duplicates of all)
• Power cables (+duplicates of all)
• Power adapter (AUS->US)
• Apple Charger
• One trick is to get rid of the standard two-prong cable, or extension from the brick, and re-use the one that comes with all the Canon battery chargers (… as long as you get a duplicate!).
• MacBook Air 13”, i7, 250GB SSD
• 1 x USB->RJ45 (Ethernet adapter for the MBA)
• 1 x iPhone cable
• Headlamp
• 3 x BL-5c, GPS Logger QStar BT-1000P – reviewed it here
• Batteries
• 4 x Canon Batteries, LP-E6
• LP-E6 Charger
• 2 x 4 x AA
• 2 x 2 x AAA
• Camera
• Canon 50mm f/1.4 USM
• Canon 17-40 f/4 USM
• Canon 70-200 f2.8 IS USM MKII — heavy!!
• Canon 5D MKII
• Canon 430EX II Flash
• Off-Shoe Flash Cord, OC-E3
• Sigh-Ray Filters
• 1 x ND-2G-HS
• 1 x ND 2G-SS
• 1 x ND-3 Rev. Graduated
• 1 x ND-3G-SS
• LEE Filter holder
• Lens cleaning solution
• Lens & Screen cloth
• LENSPEN
• Lens cleaning tissue paper
• 4 x 8GB CF, SanDisk Extreme III
• 1 x 72mm Circular Polariser, Hoya HD
• 1 x 72mm ND400, Hoya
• GorillaPod - had to leave my Manfrotto 055-XPROB at home, as I didn’t feel it could accompany me comfortably. Suggestions welcome for a light and manoeuvrable tripod for trekking
• Black Rapid RS-4 camera strap
• Book, The Last Don by Mario Puzo – during those down times such as airport transfers, and lunch breaks. Would happily replace it with a Kindle if I had one – only due to the weight of course!

Then add food, water and toiletries and you end up with a very nice ~20kg pack.